Re-Thinking Akalyptos Space, Master Thesis Project

Page 1

MASTER IN ADVANCED COMPUTATION FOR ARCHITECTURE AND DESIGN

MaCAD 2022 -

Master Thesis Project

RE-THINKING A K A L YP T O S SPACE Implementing a Machine learning Strategy for Activating the Uncovered Spaces in the Athenian Urban Blocks

2022, Printed in Barcelona, Spain

Copyright for MaCAD program at IAAC

Institut d’Arquitectura Avançada de Catalunya Pujades 102 baixos, Poble Nou, 08005 Barcelona

MASTER IN ADVANCED COMPUTATION FOR ARCHITECTURE AND DESIGN

RETHINKING “AKALYPTOS” SPACE:

Implementing a Machine learning Strategy for Activating the Uncovered Spaces in the Athenian Urban Blocks Faculty: Angelos Chronis

Jumana Hamdani & Maria Papadimitraki

TABLE OF CONTENTS 01 RESEARCH
01 | 1 Abstract…………………………………………………………………………..………07 01 | 2 Introduction………………………………………………….………………..………09 01 | 3 Research
01 | 4 State of
Art
02 TASK 02 | 1 Problem………………………………………….…………………….…………………18 02 | 2 Solution………………………………………….……………………………..………24 03 ANALYSIS 03| 1 Terminology……………………………….………………………….……………..…28 03 | 2 Urban Analysis……………………….………………………….……………….…34 03 | 3 Local Climate………………………………………….………………...………..…47 04 METHODOLOGY 04 | 1 Process………………………………………….………………………….……………60 04 | 2 Workflow……………………………………………………………………………….62 05 DATASET CREATION 05 | 1 Extract………………………………………….…………………………………………66 05 | 2 Compile……………………………………………………...…………………………70 05 | 3 Exploratory Data Analysis
05 | 4 Analysing
05 | 5 PCA
06 RESULTS 06 | 1 Clusterf-AI
06 | 2 Clusterf-AI
06 | 3 Clusterf-AI
06 | 4 Classif-AI
06 | 5 Classif-AI
06 | 6 Classif-AI
07 DEPLOYMENT 07 | 1 Evaluation………………………………………….……………………….….…..…142 07 | 2 Website………………………………………………………………………………..150 08 DISCUSSION 08 | 1 Conclusions ……………………………….…………….……..……………..……160 08 | 2 Further Steps…..…………….……………………………….………….……..…162 09 APPENDIX 09| 1 Research Wall………………………………………….………………………..…..166 09| 2 Βibliography………………………………………….…..………………….…...…168 10 ACKNOWLEDGEMENT 10| Acknowledgement………………………………………….………………….…….172
FRAMEWORK
Hypothesis………………………………….………………..…………11
the
…………………………………………….……………….…..……12
(EDA)…………………..……..………………88
Correlations….…………………………...……………………..…92
& FAMD……………………………………………...………..……………….96
Method…….…..………………….……..……………………..……110
200 blocks…..……………….……..…………..…………..……112
500 blocks…..………………….……………….…………..…..120
Method…….…….……………….………………………..…………130
200 blocks…….…………….……………..………………………132
500 blocks…….……………….…………………..…..……….…136

RESEARCH FRAMEWORK

01
01 | 1 Abstract 01 | 2 Introduction 01 | 3 Research Hypothesis 01 | 4 State of the Art

RESEARCH FRAMEWORK

Abstract

Athens, the Mediterranean metropolis, hosts the hottest summer in Europe, and in just a few decades, the city could be partially uninhabitable. As the city's newly appointed chief heat officer, Eleni Myrivili, her assignment is to protect the Greek capital from an invisible but ever-present threat, global warming. In addition, the high building and population density, and the limited number of parks and green areas, all increase the heat island effect, putting people, ecosystems, and the city at high risk. This research explores the potential of the "akalyptos" spaces in the urban blocks, the proper exploitation of which could significantly affect the city center's attractiveness, improve the microclimate and strengthen the interpersonal relationships between the residents of each urban block. The Akalyptos Spaces could be one of the possible solutions ensuring the survival of Athens. Geometric and urban features of the akalyptos can define those spaces' particular usage, privacy character and current score. In order to understand those features, we explore the application of artificial intelligence in two machine learning methods: one is clustering the data into seven subcategories, and the other is classifying the medium subareas inside the akalyptos for social inclusion, whether it is for the inhabitant of the block, the neighborhood usage, or the public. To train our machine learning model, we create a dataset from 11 neighborhoods spread around within the administrative boundary of Athens, with a range of 10-180 blocks in each neighborhood, understanding the impact of dataset variety on the effectiveness of the machine learning models. In order to evaluate the effectiveness of the dataset we generated, we trained two models for each method, one with 200 blocks with 34 features and the other with 500 blocks with 17 features. These models would be evaluated locally in Grasshopper by visualizing a parametric verified geometry and deploying the model on the akalyptos of selected blocks from Thessaloniki. The result of the best-performing model would be used in an AI web app, where the users, like residents of the city, or planners in the municipality, can be aware of the current situation and the possible transformation to make better decisions for improving those spaces.

Keywords: Athens, Akalyptos, Machine learning, Classification, Clustering, Dimensionality Reduction.

7
Image Credits https://issuu.com/panosdragonas/docs/madeinathens_catalogue

RESEARCH FRAMEWORK

INTRODUCTION

The polykatoikia (πολυκατοικία, literally "multi-residence," is a multi-story building that includes many detached apartments) was established as the dominant form of housing in Athens in the 1930s. It was initially designed by a group of young architects connected to the pioneers of the modern movement.

These early structures were commissioned by prosperous and cultured landowners who sought social cognition, and they were invented to meet the needs of that period's social and economic elite. (The structure of the typical polykatoikia had been an evolution of Le Corbusier's Dom-Ino system, with a reinforced concrete frame of columns and slabs, plus an elevator shaft and staircase that allowed for flexible floor plans).

After World War II and the Civil War that followed (1946-1949), Greece was, to a large extent, destroyed. Between 1951 and 1981, the population of Athens more than doubled, from 1,379,000 to 3,027,090. Internal immigrants moved from the rural areas to the cities causing a massive demand for housing. From 1950-80 approximately 35.000 polykatoikia were built, when up to that period in Athens, only 1.000 buildings over five floors existed. The polykatoikia very soon became the dominant building typology and led to the massive dissemination of the modern vocabulary in the Greek city.

The infinite repetition of those building typologies produced an urban landscape that covered the whole Attica basin. Every single piece of land that was possible to build on was built. Only the tops of the hills, the coasts, the archaeological sites and a few public spaces have escaped development. The rest of the landscape had been transformed into a dense, small-scale urban environment. Furthermore, architects had limited involvement in the production of polykatoikia during the post-war reconstruction. Thus, the architecture of these buildings was straightforward, with a repetition of a typical floor plan and a replication of linear balconies that formed the facade. These multi-story apartment buildings allowed population density and complementarity of uses (residential offices, while small trade, cafes, and restaurants, were hosted on the ground floor). Nowadays, most of the center of Athens and many other city quarters consist of those buildings.

9

RESEARCH HYPOTHESIS

Are there untapped urban voids that could improve the living conditions of the residents of each urban block and improve the microclimate? For example, akalyptos spaces could be a solution to the problems that the city is facing.

Akalyptos (ακάλυπτος, literally “uncovered space”) is the part of the plot that is not built due to the building regulations and the coverage factor allowing free space in the middle of the urban block. It belongs to all the owners of the apartments of the attached polykatoikia.

Despite the building regulations that define that ⁄ of the akalyptos space should be unpaved to eliminate the risk of floods, most of those spaces stay untapped and unused, creating an unpleasant environment in the middle of the urban blocks.

11
Image Credits :Arcgis

RESEARCH FRAMEWORK

STATE OF THE ART

I. Barcelona: Block City

Normally monuments are exceptions and generators of the urban plan, but in Ildefons Cerdà’s Eixample in Barcelona the dominant type of city, the urban block, subsumes any other types. For example the Sagrada Familia by Antoni Gaudi can be seen less as a monument than a church subjugated to the block structure, with its plan developed from the cloister and not a Gothic church.

Cerdà defined the blocks as pieces of public infrastructure through the concept of ‘interway’, which is distinct from the isolated block structure and islands that Rem Koolhaas discovered in Manhattan, raising the question of what a new urban idea of the interway could be.

Image Credits: Projective Cities, AA
12

Ildefons Cerdà’s Eixample in Barcelona

“Cerdà focused on key needs: chiefly, the need for sunlight, natural lighting and ventilation in homes (he was heavily influenced by the sanitarian movement), the need for greenery in people's surroundings, the need for effective waste disposal including good sewerage, and the need for seamless movement of people, goods, energy, and information.”

13
Image Credits: Four Walls: BarcBehind elona’s Lost Utopia, Failed architecture

RESEARCH FRAMEWORK

STATE OF THE ART

II. KEMP, a digital participatory design platform

Master Thesis, Students: Adamopoulos Georgios, Aronidis Elias, Pappas Phillipos

Supervisor: Papalexopoulos Dimitris, National Technical University of Athens, 2016

This thesis is based on the typical Athenian multi-storey building, “Polykatoikia”. Μore specifically, focuses on its inability to provide adequate and quality shared spaces, equally accessible to all residents. They develop and propose KEMP, a toolkit for the design of multi-level, adaptable outdoor shared spaces, positioned within the three-dimensional void of the typical Greek city block. It combines urban design and analysis tools with a communication platform for the active collaboration of residents and engineers, in the context of a digitally augmented participatory design process. Through beta version implementation of the proposed digital tools and participatory process simulation, they attempt an indicative design solution for the creation of shared spaces in an existing city block, while assessing the efficiency of these tools for augmenting the participatory design process.

Image Credits: https://dspace.lib.ntua.gr/xmlui/handle/123456789/44814?show=full
14

Image Credits: https://dspace.lib.ntua.gr/xmlui/handle/123456789/44814?show=full

15

TASK

PROBLEM/ City Scale

18
Image Credits: DW Documentary
19

TASK

PROBLEM/ City Scale

The increase of Urban settings by 2050. Urban living limits access to NATURE and can increase exposure to certain environmental hazards, such as air and noise pollution, raise temperature.

Most Asphalt Least free space

This research aims to investigate the potential of those uncovered areas. The connection of the individual akalyptos areas could create a shared space, which would be vital in the heart of every building block. It could become the place to meet, rest, and play, to get to know the neighbors, or it could simply improve the microclimate of every apartment facing it and the resilience and attractiveness of every neighborhood in the city center.

Furthermore, this city development will be based on the following principles: reconsideration, recreation, participation, and interaction of the inhabitants, following a bottom-up approach.

Using machine learning algorithms, we are attempting to classify whether the akalyptos space would be suitable as a shared free space for the residents of the urban block itself, as a shared space for the neighborhood, or lastly, applicable for public use. The classification will be based on the following factors:

● Height and uses of attached buildings

● Size and shape of “akalyptos” space

● Direct access from the street or not

● Existence of empty plots in urban blocks or not

● Uses or lack of uses of the surrounding neighborhood.

2nd highest density in EU Athens severe heat waves Green space per capita less than most EU cities
Image Credits: http://tomorrows.sgt.gr/images/e-book/Tomorrows_Book_V_E_B OOK.pdf 20

TASK

Akalyptos Space current misuse

Possible uses: private or semi-public gardens, gathering space, playground, a place for effective waste disposal, evaporative cooling potential, lending library, sporting facilities, art space, open-air cinema, theater plays, small events, small cafe, gastronomy facilities, other cultural & commercial uses. akalyptos

PROBLEM/ Neighbourhood Scale akalyptos Inefficient use lack of usesactivities Sub-areas division lack of interaction lack of greenery
21
TASK PROBLEM/ Neighbourhood Scale Akalyptos Space current misuse https://www.instagram.com/p/CWLrXTxqYb_/ https://www.athensvoice.gr/life/life-in-athens/766869/o-marios-voutsinas-fotografizei-tous-akaluptous-tis-athinas/ 22
23

TASK

SOLUTION/ Akalyptos Space

Our project comes as a solution responding to rise of temperature in the city, the lack of greeneries, public spaces, and the misused of the uncovered courtyard spaces inside the Athenian urban blocks. The Webapp would be designed for the decision makers including urban planners, architects, politicians and even inhabitants to participate with their opinions for a better usage of the Akalyptos spaces. This project will propose the optimal locations and usage of the sub areas in side the akalyptos for social inclusion.

Increase greenery/ water elements Add urban furniture Enrich outdoor activities Strengthen social relationships akalyptos Improve microclimate Increase urban attractiveness Strengthen social inclusion
sub-akalyptos in urban
24
Images Credits: Akalyptos SKG
Connect
block

TASK

SOLUTION/ AI Webapp

Researchers

Application Users

Residents

an AI webapp for residents, municipalities, and researchers to unveil the potentials of the akalyptos space

Municipalities 25

ANALYSIS

TERMINOLOGY

# Polykatoikia

Πολυκατοικία (Polykatoikia), the multi-storey apartment building, appeared as a building type in Athens all along 1910. Polykatoikia,though, as the social & legal contract which is the essence of it, was constituted and established with the 1929 law of horizontal ownership, according to which each apartment could have a different owner and the management of shared spaces is done collectively. Since then and through the system of antiparochi, polykatoikia spread out all over Attica region, taking over every free space and shaping the image of its urbanscape, as well the human geography of the modern metropolis.

The mass spread of polykatoikia was abetted by its cheap standardized construction method, the easily adaptable floor-plan which as a reinterpretation of maison dom-ino implied its infinite reproductions with small variations, its relatively small size that economically enabled its wide dissemination, and the maximization of profit due to the variety of possible uses which could co-exist in a single building: shops, offices, storage spaces, workshops, luxury last floor apartments and cheap basements, even public services.

# Akalyptos

Ακάλυπτος (akalyptos means uncovered in greek) is the mandatory empty space in the backside of polykatoikia and it is what remains of the land plot when the permitted building coverage ratio is fully exploited and built.

With the introduction of polykatoikia, back in 30s, the building coverage ratio was not yet regulated adequately by the law and it was up to each construction to find the ideal analogy, thus ending up with coverage of 80-90% or even 100% of the land plot. The 1955 law set the limit to 70% for the center of Athens, where it is till now.

28

# Buildings/ Urban Block

Building/ Urban block Οικοδομικό τετράγωνο (Oikodomiko Tetragono) is every single buildable land area, which is included between a validated street plan or between the borders of a settlement and is surrounded by common spaces. (NOK, General Building Regulation)

The First General building code designated the building block as the minimum unit of urban planning and city formation, though, in practice, it was the land plot that played that role.

the urban blocks and the street network produced, have a common typology. Its main features are their small size, their front development, their piecemeal production, their diversity and their multifunctionality as also its density and insufficient width respectively.

#Oikopedo

Οικόπεδο (Oikopedo) is the land plot in the urban blocks. Land plots, especially in the center of the city are generally quite small since the fragmentation of the land was encouraged by the state as another policy to create extra value out of land without spending any money.

Whereas the building block is set out by the law as the elementary unit of urban planning, it was finally the land plot that proved out to be the module which defined the modern cityscape of Athens. Due to its characteristics, small, overbuilt and overcrowded, it contributed to this fetching urban diversity of Athens,

29

ANALYSIS

TERMINOLOGY

though not without problems.

The first Building Code of 1929 allowed for new constructions even in plots that gave a minimum floor area of 40 sq. Fragmentation of land combined with high floor area ratio and high building coverage ratio, which reached even 80%-90% during 30s when the building regulation was quite relaxed, squeezed any free –green space out of the land plot.

# Aniparoxi

The Αντιπαροχή (antiparochi) system was enabled by the 1929 law of horizontal ownership and according to it a landowner could turn over the plot to a constructor, usually a small scale construction company, in order to build a multi-storey apartment block receiving in exchange an agreed number of apartments in the finished building.

Antiparochi became the synonym of Reconstruction Era of postwar Greece and the Athenian polykatoikia- 90% of them were built with the system of antiparochi. The great interior migration wave plus the casualties of the war, counting thousands of demolished houses, urged for a fast mass housing program.

Antiparochi was a kind of a social welfare policy of the State, which couldn’t afford to finance directly a social housing program so it invested in this type of semi-informal urbanism, creating favorable conditions for the unplanned expansion of the city. This was also a State directive to incentivize and promote the construction sector as its basic productive activity that in the following decades leveraged the Greek economy. Tolerant as it was with illegal

30

building, and very accommodating with building quality controls, the Greek state furthered primed new constructions imposing a high tax on property transfers as an anti-incentive to buying existing buildings.

The mechanisms of capitalizing on land were to raise the floor area ratio thus creating surplus value out of the existing land, against the urban space quality and public space. There were special tax incentives for antiparochi that enabled its wide spread out – it was almost free of tax for the land- owner, who had just to pay a very small fee (around 3000 euro).

After 2006 the law on antiparochi changed and now the owner of the land has to pay a tax of 18% on the value of the apartments he receives as antiparochi, thus putting an end to the golden era of antiparochi

Definition Source: https://cohabathens.org/portfolio/oikopedo-land-plot/

31
Images Credits: Venice Biennale 2012: Greek Pavilion

ANALYSIS TERMINOLOGY

#Retire

Ρετιρέ (Retire) from the French word retiré meaning withdrawn is called the apartment whose outer wall is inside that of the facade of the rest of the building. Retire apartments are located in the last floors of the polykatoikia and have larger balconies than those of the rest of the floors, which are called verandas.

The fact that they are located in the last floors and that they provide larger outdoor space, makes them more sought after and thus more expensive in comparison to the rest of the apartments. The retire apartment are also determined by the law.

Since the period of the great urbanisation in Greece in the late ‘60s and ‘70s, the ownership of a retire apartment was a sign of urban affluence and that is still the case today.

32
Credits: https://www.archetype.gr/blog/arthro/i-polikatikia-os-mithologia
Images

Retire

Akalyptos Polykatoikia

Urban block

Images Credits: http://www.arch.ntua.gr/index.php/ergasia/1-polykatoikia2/ Image editing for the purpose of current project

33

ANALYSIS

From The Decentralized Administration of Attica, we focused on the administrative boundary of the city of Athens. Then, we chose 11 neighborhoods spread around the city with a range of 10 - 180 blocks in each neighborhood, understanding the Impact of dataset variety on the effectiveness of the machine learning models.

35
ANALYSIS URBAN ANALYSIS/ Selection Sub-areas KOLONAKI NEAPOLI EXARCHIA GIZI KYPSELI AGIOS NIKOLAOS 36
PLATEIA AMERIKIS
AMPELOKIPOI NEOS KOSMOS 37
VOTANIKOS KOUKAKI

ANALYSIS

URBAN ANALYSIS/ Selection Sub-areas

Sub-areas Selection

We focused on 11 Neighbourhoods within the administrative boundary of the city of Athens. Each district of the Athenian city center has its own distinct personality. Therefore, the neighborhood selection was based on the different characteristics, such as the urban and natural environment, the residents' profile, and the activities each area has. We also chose Urban blocks and Akalyptos within these areas with diverse shapes, sizes, and closedness/ openness to have an equally represented dataset.

The main characteristics of the selected neighborhoods are the following:

38
39 Images Credits: https://www.instagram.com/p/Ch2UO8aLrVH/

ANALYSIS

URBAN ANALYSIS/ Selection Sub-areas

Kypseli: multicultural and artistic area, a remarkable mix of architectural styles, from crumbling neoclassical mansions to striking Bauhaus designs, eateries from all over the world, bars and cafes, numerous theaters, shops and events, municipal market, Pedion Areos Park.

40

Images Credits: https://www.in.gr/2022/08/01/greece/allazei-eikona-o-lofos-tou-strefi/

Exarchia: revolutionary district, place of social, political, and artistic fermentation, multicultural ambiance, students, academics, and intellectuals, alternative side of Athens, bustling nightlife, small cafes and quaint taverns, theatrical plays, festivals, various events, vintage shops, street art, National Archaeological Museum of Athens, National Technical University of Athens, Lofos Strefi

https://athensbylocals.com/exarchia-neighborhood-in-athens/ 41

URBAN ANALYSIS/ Selection Sub-areas ANALYSIS

Votanikos: is named after a nearby botanical garden situated southwest (Athens Botanical Gardens). The eastern part is residential, and the western part is forested and industrialized.

Gizi: residential area, hospitals, country court, small cafes and restaurants Pedion Areos Park, Finopoulou Hill, eight blocks old Bauhaus style refugee Houses of Alexandras Avenue constructed in the 1930s

42

Kolonaki: wealthy, chic, and upmarket district, expensive commercial shops, high-end boutiques, popular cafes, modern bars and restaurants, lively atmosphere and nightlife, museums, embassies, public buildings, Lycabettus hill.

Koukaki: eclectic and charming district, beneath the shadow of the Acropolis, residential area, Athenian cultural hub, art galleries, themed coffee shops, international eateries, antique stores, hotels, Filopappou Hill.

43

URBAN ANALYSIS/ Selection Sub-areas ANALYSIS

Neos Kosmos:

name meaning “New World,” held out the promise of a new beginning for refugees from Asia Minor, with many residential buildings, small businesses, restaurants, bars and cafes, live music stages, theaters, Onassis Stegi cultural center, theatrical and musical performances, art exhibits, festivals and discussions, National Museum of Contemporary Art.

Plateia Amerikis:

took its name from America Square, around which it was developed. It began to develop rapidly in the early 1950s when notable high-profile and high-quality apartment buildings were built around the square. It was one of the most prestigious districts of Athens, mainly a residential area with small restaurants and cafes.

44

Nikolaos: vibrant residential and commercial neighborhood, a wide range of stores, cafes, and eateries, 19th-century Agios Nikolaos Church, Byzantine Cathedral of the Holy Trinity.

Agios

Neapoli: a mainly residential area that connects the two most controversial districts of Athens, Exarchia and Kolonaki, the place that still holds “neighborhood smells” in the city center. Many students and artists were sloppy with numerous stairs on the roads of Lycabettus and Strefi Hill.

45

URBAN ANALYSIS/ Selection Sub-areas ANALYSIS

Ampelokipoi: name meaning 'vineyards,' formerly filled with strims and fields, currently one of the most densely populated districts, many hospitals, and office buildings, the only two Skyscraper of Athens are located here, Panathinaikos Football Stadium.

46

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

47

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

48

# Sun path for Solstice Days

n the diagrams we can see the sun path during the summer (21st of June) and the winter solstice (21st of December). The colours inside the sunspheres show in the 1st diagram the actual amount of global solar radiation. The darker the colour is, the less solar radiation is received during that particular hour of the day.

In the second diagram the colours inside the sunspheres show the air temperature. The higher air temperatures are observed in the summertime and from the middle of the day till the evening (west direction).

summer solstice winter solstice

21st of December & 21st of June

Sunpath & Global Solar Radiation

summer solstice winter solstice

21st of December & 21st of June

Sunpath & Air Temperature

49

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

Air Temperature

The first diagram shows that the highest temperature throughout the year is ca. 37°C and the lowest 5.5°C. The hottest period is from May till the beginning of October, while the highest temperatures appear at the beginning of June till the end of September. On the other hand, the coldest months are December to March, with the coldest temperatures observed from the end of December till the middle of February.

Regarding the hours of the day, the hottest period is seen between 09:00 to 19:00.

According to the second and third diagrams, we understand that heating for the thermal comfort of the residents will be needed from November to April, while cooling systems probably will be needed in July and August, with the hottest hours occurring from the middle of the day till the evening.

50
51

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

Relative Humidity

According to the diagrams, the highest humidity is observed in the months 52

# Sky Cover

According to the diagrams, the months with the less sky cover are from

53

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

54
55

ANALYSIS

LOCAL CLIMATE ANALYSIS/ Athens

Wind Speed and Direction

From the wind underneath, we can understand that the strongest winds occur from the north. The direction is 0 to 50 degrees from October to March, while from June to September, the main wind direction is the absolute north. The wind has a positive effect during the cooling period (passive cooling) and a negative effect during the heating period.

56

annual

57

METHODOLOGY PROCESS

Our machine learning process is defined in five steps: collecting the dataset, preparing the data, choosing the ML models, evaluating the results, and deploying the model in an AI Web app.

Extracting heights and trees features in selected urban blocks Google Earth

Urban analysis of athens & addition features in selected urban blocks Open Street Maps

creation of initial dataset 200 blocks

DATA SOURCES

FEATURES ML

machine learning Clustering 500 blocks

RESULTS

Grasshopper generate area of "akalyptos" & additional features web application of "akalyptos"

creation of more dataset 500 blocks with less features according to pca/ famd analysis machine learning Classification 500 blocks Compare Classification Results

machine learning Classification 200 blocks machine learning Clustering 200 blocks Compare Classification Results
60

To create a machine learning model that can find correct patterns, we collected reliable data on Akalyptos spaces from existing blocks in Athens. We created a dataset from 11 neighborhoods spread around the administrative boundary of Athens, with a range of 10-180 blocks in each neighborhood, understanding the impact of dataset variety on the effectiveness of the machine learning models. We used Google Earth to extract the buildings' heights, the roofs' material and shape, the number of trees, the distance to the closest parks, and the distance to the city center, which is the commercial triangle of Athens. From Open Street Maps, we collected all the urban features that play a role in shaping the model. These features are the coordination points, number of buildings, point of interest, average heights of the buildings, number of pools, street types, average speed, and number of parking within the one block. Then we did the same for a 200m radius from the center of each Akalyptos to gather average heights of buildings, point of interest, transportation points, street type, average speed, neighborhood type, and the number of parking. In Grasshopper and Rhino, we drew 500 Akalyptos outline shapes and their blocks. We imported the data from OSM with the ELK plugin in Grasshopper to generate the buildings in 3D; we calculated the Akalyptos area, size, Block area, unpaved ratio, and built ratio to block surface. With Ladybug, we did the environmental analysis. We calculated the Akalyptos sun hours for the longest day and sun hours for the shortest day, as we got the block's deviation axis and its orientation. In Grasshopper, we did a script based on research on defining the Akalyptos in sub-areas with different sizes for different uses. So, we generated three sub-areas one Large for environmental purposes, a medium size for the social interaction of the residents, and a small size for private uses. We extracted the area large and medium to include them in our dataset, as well as the evaluation of each large area if they fulfill the ⅔ regulation of the unpaved surfaces from the municipality of Athens. In the beginning, we collected 200 blocks with 34 features; then, we wanted to increase the data. We collected 300 more, 500 blocks in total with 19 features instead. In the second step of preparing the data, we gathered all our datasets and started by cleaning and removing unwanted data, missing values, rows, columns, duplication values, data type conversion, etc. Later, we split the data into two sets: a training set for Athens and a testing set for Thessaloniki.

In the third step of choosing the ML model and in order to understand those features, we explore the application of artificial intelligence in two machine learning methods: one is clustering the data into seven subcategories, and the other is classifying the medium sub-areas inside the akalyptos for social inclusion, whether it is for the inhabitant of the block, the neighborhood usage, or the public. To compare the results correctly, we run a clustering model o the 200 blocks and another on the 500 blocks and evaluate which performed better. For the classification: we run three shallow learning models: Logistic regression, XGboost, and Random first classifier, and six deep learning models ANN with different layers architecture, and since we have to do that process on the large dataset and the small one, we ended up running 12 classification models in total.

In the fourth step, these models were evaluated locally in Grasshopper by visualizing a parametric verified geometry and deploying the model on the Akalyptos of selected blocks from Thessaloniki. Finally, the result of the best-performing model would be used in an AI web app, where the users, like residents of the city, or planners in the municipality, can be aware of the current situation and the possible transformation to make better decisions for improving those spaces.

61

METHODOLOGY

WORKFLOW

62
63

EXTRACT

Open Street Maps

Rhino /Grasshopper

Google Earth

EPSG.io

Expert Input

Contribution to OSM Data

Subarea unpaved

Large Subarea

Unpaved Ratio

Axis Angle Axis Orientation

Akalyptos Sun Short hours

Akalyptos Sun long hours

Medium Subarea

Subarea Evaluation 2/3

Built ratio to block surface

Akalyptos Size

Akalyptos Area

Block Area

Distance to Park

Number of Parking/block Distance to City Center

34 FEATURES Number of 200m neighborhood

Final AKALYPTOS

DATASET CREATION/ 200 Blocks
66

Akalyptos Entrances Center/ Polygon coordination

Akalyptos Shape

Number of buildings/block

Point of Interest/block

FEATURES

Average height/block

Number of Trees/Akalyptos

Number of Pools/Akalyptos or roofs

Street Type/Block

Average Speed/Block

Average height/200m neighborhood

Point of Interest/200m neighborhood

Transportation points/200m neighborhood Street Type/200m neighborhood

Average Speed/ 200m neighborhood

Neighborhood Type/ 200m neighborhood Parking/

Final
AKALYPTOS
neighborhood Orientation 67
DATASET CREATION/ 500 Blocks EXTRACT Final AKALYPTOS 19 FEATURES Distance to City Center Distance to Park Akalyptos Area Akalyptos Size Medium Subarea Open Street Maps Rhino /Grasshopper Google Earth EPSG.io Expert Input Contribution to OSM Data Akalyptos Sun long hours Axis Angle Axis Orientation Unpaved Ratio 2/3 Built ratio to block surface Block Area Number of Parking/block 68

Akalyptos Entrances Center/ Polygon coordination

Akalyptos Shape

FEATURES

Average height/block

Number of Trees/Akalyptos

Number of Pools/Akalyptos

Point of Interest/200m neighborhood

Final
AKALYPTOS
Orientation 69

DATASET CREATION COMPILE

Building heights, roof material and shape updated in OSM Buildings

Average heights of Area

Building
detailed
heights
70

Trees inside Akalyptos/ Buildings Height of the Block

The accuracy of the AI model is directly correlated to the quality of data used to train it. For example, in the Akalyptos Typologies, we labeled three categories: Size, Shape, and number of entrances into the akalyptos. In the data preparation workflow, we wanted to start building reliable AI models based on clean data. Therefore, two main features were taken into consideration: the number of trees in the Akalyptos, as it plays a massive role in the environmental analysis of the uncovered spaces, and second is the height of the buildings as they affect the light, shadow, and wind circulation inside the akalyptos. We observed on google earth the numbers of trees and applied them to the OSM data repository. For the height, we subtracted the value on the top of the roof and the street level value, giving us the total height of each building. According to our observations, we also defined whether the uncovered space is used as parking or whether there are water features (such as fountains or pools) in the urban blocks' open spaces and added this information to OSM data.

● Akalyptos Trees 633 ● Building heights 2549 ● Roof shape 2549 ● Roof Materials 2549 ● Pools 37 ● Parking Lots 50 Observe Trees/ Water Features/ Parking Lots & Calculate Buildings heights Edit OSM Open Street Maps Google Earth 8367 contributed features to OSM
71

DATASET CREATION

Akalyptos Space sizes & shapes

COMPILE Images Credits: https://www.instagram.com/p/CWLrXTxqYb_/ 72
73

DATASET CREATION COMPILE

Akalyptos Typologies

74

Buildings Akalyptos

75

Typologies Size

Typologies shape Typologies closedness

Polarised center

The accuracy of the AI model is directly correlated to the quality of data used to train it. For example, in the Akalyptos Typologies, we labeled three categories: Size, Shape, and number of entrances into the akalyptos.

For the size of the Akalyptos: we labeled them based on the area Xsmall for less than 300 sqm, Small between 300-600 sqm, Medium between 600-900 sqm, Large between 900-1200 sqm, and Xlarge any area over 1200 sqm.

For the shape: if the akalyptos is narrow, it is Linear, and if it has divided into more than one courtyard, it is Disconnected, and if it has centered with a wide proportion area, it is a Polarised center.

For the closedness/openness of the Akalyptos, if it has no entrance, it is Enclosed, we label it 0, if it is with one opening, we label it 1, and if it is two or more entrances, we label it with the number of entrances it has.

DATASET CREATION
XS
S
● 4.
● 5. XL
● Enclosed ● One opening ● more ≥ ● Linear ● Disconnected ●
COMPILE ●
(≤300 sq) ●
(300-600 sq) ● M (600-900 sq)
L (900-1200 sq)
(>1200)
76
Strefi Hill M 600-900 sq S 300-600 sq XS 0-300 sq XL >1.200 sq L 900-1.200 sq Akalyptos Typologies 77

DATASET CREATION

COMPILE

RATIO BUILT-UP AREA /URBAN BLOCK AREA

Akalyptos Built-up Area Urban Block

Strefi Hill

78
from North 79 27-36 degrees 18-27 degrees 9-18 degrees 45-54 degrees
Deviation
Strefi Hill

DATASET CREATION COMPILE

Distance to Park/ Hill

Walk Distance to Park/ Hill
80

Distance to Commercial Triangle & Main Metro stations

Walk Distance to Main Squares & Metro Stations of Commercial Triangle

81
CREATION Average Sunlight hours: 9.771768 COMPILE Solar Analysis/ Summer Solstice
82
DATASET
Sun-path summer

Solar Analysis/ Winter Solstice

Sun-path winter

Average Sunlight hours: 3.121372
83

DATASET CREATION

Akalyptos Social Subareas Labeling COMPILE

Finding perimeters with shape "Π"

Three consecutive sides which make an internal turn

1 2 5 6

Determination of perimeter cavities

Generating sub-areas for the perimeter cavities

Unifying internal sub-areas

Generating out of the multiple intersected circles of one category unified sub-areas

Smoothing of shapes

Generating one smooth shape for each sub-areas

84

Finding nodal points on the median axis

Points of the central axis which are terminations and do not fall into the perimeter cavities

Determination of internal sub-areas

Connection of circles centered on the nodal points of the median axis and with radius their distance from the boundaries of the akalyptos

Graph Generation

As nodes are defined the sub-areas centers and as edges the connections of the centers

Finding degree of adjacency

Identifying with how many nodes each nodal point is in direct contact with (topological distance =1)

3 4 7 8
85 3 3 3
3 3 3 3

DATASET CREATION

Akalyptos Social Subareas Labeling COMPILE

To create the subareas in the akalyptos, we based our work on a project thesis called Kemp, a digital participatory design tool for the Athenian blocks. They identify the small pockets of the akalyptos, which is what most of the pink area will be for private use. The shortest distance across the network creates joints. In these joints, circles would be generated with the shortest path radius to the boundary of the akalyptos. The result would be a variety of different sizes of circles intersecting together. We classify them into three categories: pink for private use and they all usually in the narrow edges, yellow for social inclusion as the medium sizes, and the most prominent circles become the green zones for the environmental aspects, to be either planted, added water features to improve the microclimate in the akalyptos space.

86

Sub-area large → Environmental Purposes/ Common

Sub-area medium → Social Interaction/ Semi-common

Sub-area small → Private Space

Strefi Hill

87

DATASET CREATION

EXPLORATORY

DATA ANALYSIS (EDA) 200 - 34 FEATURES

Pairs plots are a powerful tool to quickly explore distributions and relationships in a dataset to find patterns, relationships, or anomalies to inform our subsequent analysis. The Residents' sub-area generally represented the most in the dark purple color and has a more significant relationship with certain factors than others. For example, it positively correlates with the number of buildings, average heights, and distance to the park and city center. In contrast, it has a negative relationship with Akalyptos closedness, number of points of interest, number of parkings, and pools in the block.

88

The histogram plots show an approximate representation of the distribution of our numerical data. For example the numbers of pools in a block or the number of parking are not well distributed in our dataset, which means in the future it’s better include blocks that has more of these features to balance the dataset. While the average height and the distance to the city center features are well balanced.

89

DATASET CREATION

EXPLORATORY DATA

ANALYSIS (EDA) 500 - 19 FEATURES

We did the same for the 500 blocks but with fewer features. Here we observed that the Residents' sub-area still represents the most in the cyan color and has a greater relationship with certain factors than others. For example, it has a positive relationship with the distance to the park and center, while it has a negative relationship with the Akalyptos closedness, the number of pools, and point of interest in the block.

90

In the histogram plots, although we increased the dataset with 300 blocks more, still we observe uneven distribution of some features in the dataset, like public point of interest in the blocks, and the numbers of pools. While the distance to the park and the commercial triangle, show better distributions. For most of the features this is something we also expected, since the blocks that have for instance pools are very few in comparison to those that do not.

91

DATASET CREATION

ANALYSING CORRELATIONS HEATMAP

200

In the Heatmap of Pearson correlation coefficient matrix, where we can observe a graphical representation of the correlation between the different variables, we noticed that the ratio of the built area to the total surface of the block is highly correlated with the Akalyptos area in the 200 blocks/34 features. In contrast, the Large Subarea features negatively correlate with the Akalyptos area and the unpaved ratio feature.

We also observed a high proportionality between the ratio of the built-up area with both the numbers of trees and the Akalyptos closedness.

92
93

DATASET CREATION

ANALYSING CORRELATIONS HEATMAP 500

In the 500 blocks/ 19 features Heatmap, we noticed that the ratio of the built area to the block surface is still highly correlated with the number of trees, the Akalyptos closedness, the Akalyptos unpaved ratio, and the classification medium area. That leads to considering the reduction of these features, which might disrupt the relationship among the variable evenly.

94
95

DATASET CREATION

PCA & FAMD 200 Blocks

Essential Features

In the following image, we can observe the most essential contributed feature to each principal component. For instance, the most critical contributed feature for Principal Component 0 (PC0) is the unpaved_ratio of Akalyptos space.

Line Plot_ Variance Explained

We used the line plot to understand how many Principal Components we should retain in our training process to have a good representation of the data without adding unnecessary noise and thus to have a well-balanced dataset.

The orange line represents the cumulative sums or running totals used to display the total sum of data as it grows over time.

By plotting the "eigenvalues" on the line plot, we can observe that the first 12 PCs show the highest variation. PC0 explains about 20% of the data variance, the next one up to about 35%, and PC11 does not explain that much more. The purple line shows the remaining percentage (opposite representation of cumulative sums). In conclusion, we can remove PC11 to PC21.

PCA works only with numerical data. We could use the representation we get from our principal components analysis as our features because, technically, it is creating new features for us that combine different features with different weights.

96
20% 35% 45% 55% 63% 68% 73% 90% 80% 78% 97

DATASET CREATION

PCA & FAMD 200 Blocks

Biplot Principal Component (PCA)

We used Biplot for the following reasons:

● Find the main vectors of variation in our data. PCs are new variables that are a linear addition or factorization of the other variables, maximizing the variance.

● In PCA, a representation system is built centered at the intersection between the components, and the components become the axes. It is a flattened representation of our multidimensional data.

● The way the PCs are sorted is by their importance (i.e., PC0 explains much more variability in our dataset than PC3). If we want to compress our data without losing information, we should use the first ones that explain the most out of our data.

● The first two PCs will be most likely orthogonal (this is the case also in the graph on the next page) because they point in different ways (conversance-wise) and are not likely to be correlated.

● We can remove the PCs that are pointing in very similar directions (almost parallel with each other) in order to achieve dimensionality reduction.

● They are negatively correlated when diverging and forming a large angle (close to 180°).

Key Features

PC16 akalyptos area

PC17 ratio built to total surface block

PC9 number point of interest area 200m

PC2 number point of interest block

PC4 number trees block

PC14 distance to park in km

98
99

DATASET CREATION

PCA & FAMD 200 Blocks

Principal Component Heatmap

With the Principal Component Heatmap we can:

● Understand the relationship between different principal components and input parameters and also figure out which feature informs the most each principal component

● For instance the block area is informing the most PC8

100

FAMD

In the FAMD (Factor Analysis of Mixed Data), we can include both Numerical and Categorical input features.

In the first graph, we can observe a pattern using color hue and the classification of the medium areas of the akalyptos. We can see three clusters, as many as the categories of the classification of the areas.

101

DATASET CREATION

PCA & FAMD 500 Blocks

Essential Features

The following image shows the most important contributed a feature to each principal component. For instance, for Principal Component 0 (PC0), the most essential contributed feature is the unpaved ratio of Akalyptos space.

Line Plot_ Variance Explained

As explained before, we used the line plot to understand how many Principal Components we should retain in our training process to have a good representation of the data without adding unnecessary noise and thus to have a well-balanced dataset.

The cyan line represents the cumulative sums or running totals used to display the total sum of data as it grows over time.

By plotting the "eigenvalues" on the line plot, we can observe that the first 8 PCs show the highest variation. PC0 explains about 25% of the data variance, the next up to about 40%, and PC8 does not explain much more. The blue line shows the remaining percentage (opposite representation of cumulative sums). In conclusion, we can remove PC8 to PC12

PCA works only with numerical data. We could use the representation we get from our principal components analysis as our features because, technically, it is creating new features for us that combine different features with different weights.

102
25% 40% 50% 60% 65% 80% 70% 90% 103

DATASET CREATION

PCA & FAMD 500 Blocks

Biplot Principal Component (PCA)

We used Biplot for the following reasons:

● Find the main vectors of variation in our data. PCs are new variables that are a linear addition or factorization of the other variables, maximizing the variance.

● In PCA, a representation system is built centered at the intersection between the components, and the components become the axes. It is a flattened representation of our multidimensional data.

● The way the PCs are sorted is by their importance (i.e., PC0 explains much more variability in our dataset than PC3). If we want to compress our data without losing information, we should use the first ones that explain the most out of our data.

● The first two PCs will be most likely orthogonal (as in the graph on the next page) because they point in different ways (conversance-wise) and are not likely to be correlated.

● We can remove the PCs that are pointing in very similar directions (almost parallel with each other) in order to achieve dimensionality reduction.

● They are negatively correlated when diverging and forming a large angle (close to 180°).

Key Features

PCO unpaved ratio akalyptos

PC2 average buildings height block

PC4 area block

PC6 sunlight hours summer solstice

PC8 akalyptos closedness

104
105

DATASET CREATION

PCA & FAMD 500 Blocks

Principal Component Heatmap

● With the Principal Component Heatmap, we can:

● Understand the relationship between different principal components and input parameters and also figure out which feature informs the most each principal component

● For instance, the block area is informing the most PC5

106

FAMD

As we explained in the FAMD (Factor Analysis of Mixed Data), we can include both Numerical and Categorical input features.

Despite our excellent results in the FAMD Analysis of the 200 blocks, we cannot see apparent patterns in the analysis of the 500 blocks.

107

RESULTS

CLUSTERF-AI METHOD/ K-MEANS CLUSTERING

The method we selected to cluster/ group similar data is K-Means Clustering.

K-means clustering is one of the simplest and most popular unsupervised machine learning algorithms. Unsupervised algorithms make inferences from datasets using only input vectors without referring to known or labeled outcomes.

K-means aims to group similar data points together and discover underlying patterns. To achieve this objective, K-means looks for a fixed number (k) of clusters in a dataset.

A target number k needs to be defined, which refers to the number of centroids needed in the dataset. The number of centroids is an imaginary or actual location representing the cluster's center. K-means algorithm allocates every data point to the nearest cluster while keeping the centroids as small as possible.

METHOD NAME PARAMETERS SCALABILITY USE CASE GEOMETRY (METRIC USED)

K-Means

number of clusters

Very large n_samples, medium n_clusters with MiniBatch code

General-purpose, even cluster size, flat geometry, not too many clusters, inductive

Distances between points

Source: https://scikit-learn.org/stable/modules/clustering.html

110

The process followed in K-means clustering is the following:

1. Initial centroids are selected as many as inputs numbers

2. Each sample is assigned to its nearest centroid

3. New centroids are formed by taking the average of all the samples

➔ The algorithms repeat the last two steps until the difference between iterations is close enough.

K-means clustering on the digits dataset (PCA-reduced data)

Centroids are marked with white cross

Image credits: https://scikit-learn.org/stable/auto_examples/cluster/plot_kmeans_digits.html#sphx-glr-auto-examples-cluster -plot-kmeans-digits-py

111

RESULTS

CLUSTERF-AI 200 blocks - 34 features

Regardings the correlations of the input features of the 200 block, we can see from the Pearson correlation coefficient heatmap the followings:

The highest correlation we can observe between the sunlight hours of the longest and the shortest days of the year with 97% and the akalyptos area and the akalyptos size typology with 89%.

We can also see that the average height of the buildings in the block and the average height of the 200m show a high correlation (76%). In contrast, the distance to the park and the distance to the commercial triangle show a 72% correlation.

112

Furthermore, the Akalyptos size type has a 65% correlation with the number of buildings in the block, and the park distance and the average speed of the surrounding area of 200m show a 64% correlation.

Lastly, the Akalyptos area and the number of trees in the block have a 63% correlation.

As a next step, we defined how many clusters we should divide our dataset. According to the graph Distortion Score Elbow for K-means Clustering, we divided our dataset into 10 clusters.

113

RESULTS

CLUSTERF-AI 200 blocks - 34 features

Clusters Description

0: akalyptos size: S-ΧL (>300 sq), aka_closeness: 0-3, akalyptos shape: diverse, park distance: 0.19-2.2km, aver. height buildings block: 10.42-20.33m, commercial triangle distance: 1.0-3.3km, trees: 2-11, pools: 0, poi block: 0-7, num_poi_area_200m: 25-167, num_parking: 0-4, num_parking_area_200m: 0-12, ratio_built_totalSurfBlock: 62-83%, classification_areas_m: 1-3, sunlighthours_longday: 0-13.96, longitudinal_Axis_orient: 1.5-46 degrees, num_buildings_block: 7-25, av_speed_block: 18-40, av_speed_area_200m: 19.73-52.5, num_transp_area_200m: 0-13

1: akalyptos size: XS-M (0-900 sq), aka_closeness:0-2, akalyptos shape: diverse, park distance: 0.45-2.1km, aver. height buildings block: 13.8-21.8m, commercial triangle distance: 2.1-3.3km, trees: 0-6, pools: 0-2, poi block: 0-6, num_poi_area_200m: 25-144, num parking: 0-1, num_parking_area_200m: 0-16, ratio_built_totalSurfBlock: 76-93%, classification_areas_m: 1-3, sunlighthours_longday: 0-13.4, longitudinal_Axis_orient: 1.7-49.7 degrees, num_buildings_block: 5-18, av_speed_block: 21-50, av_speed_area_200m: 26-56, num_transp_area_200m: 0-14

2: akalyptos size: XS-L (0-1200 sq), aka_closeness:0-2, akalyptos shape: diverse, park distance: 0.3-1.5km, aver. height buildings block: 14-24.3m, commercial triangle distance: 0.6-3.2km, trees: 0-6, pools: 0, poi block: 0-18, num_poi_area_200m: 40-260, num parking: 0-2, num_parking_area_200m: 1-25, ratio_built_totalSurfBlock: 74-94%, classification_areas_m: 1-2, sunlighthours_longday: 0-12.8, longitudinal_Axis_orient: 8.9-54.7 degrees, num_buildings_block: 5-18, av_speed_block: 10-43, av_speed_area_200m: 10-40, num_transp_area_200m: 0-23

114

3: akalyptos size: S-XL (>300 sq), aka_closeness:0-2, akalyptos shape: diverse, park distance: 0.18-0.5km, aver. height buildings block: 9.7-19.7m, commercial triangle distance: 0.7-1.0km, trees: 0-6, pools: 0, poi block: 2-13, num_poi_area_200m: 88-279, num parking: 0-3, num_parking_area_200m: 6-15, ratio_built_totalSurfBlock: 73-85%, classification_areas_m: 1-3, sunlighthours_longday: 0-13.3, longitudinal_Axis_orient: 33.5-34.0 degrees, num_buildings_block: 11-24, av_speed_block: 10-37, av_speed_area_200m: 15-20, num_transp_area_200m: 1-4

4: akalyptos size: XS-S, XL (0-600 sq, >1200sq), aka_closeness:0-1, akalyptos shape: diverse, park distance: 0.04-1.5km, aver. height buildings block: 10.4-17.2m, commercial triangle distance: 0.7-2.8km, trees: 0-6, pools: 0-1, poi block: 0-3, num_poi_area_200m: 26-83, num parking: 0, num_parking_area_200m: 0-7, ratio_built_totalSurfBlock: 71-90%, classification_areas_m: 1-2, sunlighthours_longday: 7-14, longitudinal_Axis_orient: 9.8-47 degrees, num_buildings_block: 5-16, av_speed_block: 10-50, av_speed__area_200m: 18-45, num_transp_area_200m: 0-11

5: akalyptos size: XS-L (300-1200 sq), aka_closeness:0-1, akalyptos shape: diverse, park distance: 0.03-0.65km, aver. height buildings block: 12.3-21.5m, commercial triangle distance: 0.9-2.5km, trees: 0-6, pools: 0-2, poi block: 0-5, num_poi_area_200m: 33-137, num parking: 0-3, num_parking_area_200m: 0-4, ratio_built_totalSurfBlock: 80-94%, classification_areas_m: 1-2, sunlighthours_longday: 0-9.13, longitudinal_Axis_orient: 10.4-32.3 degrees, num_buildings_block: 3-20, av_speed_block: 18-50, av_speed__area_200m: 20-47, num_transp_area_200m: 1-13

115

RESULTS

CLUSTERF-AI 200 blocks - 34 features

6: akalyptos size: S-XL (>300 sq), aka_closeness:0-2, akalyptos shape: diverse, park distance: 0.25-1.5km, aver. height buildings block: 17.3-27.5m, commercial triangle distance: 0.5-3.2km, trees: 0-8, pools: 0-2, poi block: 4-21, num_poi_area_200m: 105-325, num parking: 0, num_parking_area_200m: 1-14, ratio_built_totalSurfBlock: 65-90%, classification_areas_m: 1-3, sunlighthours_longday: 0-13, longitudinal_Axis_orient: 7.6-36.2 degrees, num_buildings_block: 8-23, av_speed_block: 17-64, av_speed__area_200m: 18-39, num_transp_area_200m: 3-27

7: akalyptos size: L-XL (>900 sq), aka_closeness:1-2, akalyptos shape: diverse, park distance: 1.0-1.3km, aver. height buildings block: 18.5-20.4m, commercial triangle distance: 2.9-3.2km, trees: 2-16, pools: 0, poi block: 10-24, num_poi_area_200m: 128-212, num parking: 6-8, num_parking_area_200m: 9-10, ratio_built_totalSurfBlock: 71-84%, classification_areas_m: 1-3, sunlighthours_longday: 7.5-10.7, longitudinal_Axis_orient: 11.4-43.8 degrees, num_buildings_block: 17-20, av_speed_block: 33-67, av_speed__area_200m: 27-35, num_transp_area_200m: 4-8

8: akalyptos size: XS (≤300 sq), aka_closeness:0, akalyptos shape: polarized_center, park distance: 0.2km, aver. height buildings block: 20m, commercial triangle distance: 0.9km, trees: 0, pools: 0, poi block: 1, num_poi_area_200m: 146, num parking: 0, num_parking_area_200m: 2, ratio_built_totalSurfBlock: 99,96%, classification_areas_m: 1, sunlighthours_longday: 0, longitudinal_Axis_orient: 13 degrees, num_buildings_block: 6, av_speed_block: 30, av_speed__area_200m: 26, num_transp_area_200m: 12

9: akalyptos size: XS-M, XL (0-900 sq, >1200sq), aka_closeness:0-3, akalyptos shape: diverse, park distance: 0.06-0.35km, aver. height buildings block: 10.1-16.1m, commercial triangle distance: 1.2-1.5km, trees: 2-11, pools: 0-6, poi block: 0-6, num_poi_area_200m: 36-168, num parking: 0-1, num_parking_area_200m: 0-3, ratio_built_totalSurfBlock: 80-95%, classification_areas_m: 1-2, sunlighthours_longday: 0-7, longitudinal_Axis_orient: 12.7-60.4 degrees, num_buildings_block: 7-17, av_speed_block: 0-43, av_speed__area_200m: 17-32, num_transp_area_200m: 3-10

116

NUMBER CLUSTER EVALUATION DESCRIPTION

environmentally good

environmentally ok, many water features

environmentally sound, many POI, Akalyptos not ideal for public use, excellent transportation very close to the park and commercial triangle, many POI, tranquil areas with low speed akalyptos are not ideal for public use, few pools very close to the park, not ideal for public use, environmentally worse than 2, but more luxurious

environmentally sound, many trees

many trees, no pools, many POI, less quiet area, high speed, far from commercial triangle worse best most luxurious and closer to park, low speed

3 2 9 4 6 7 5 8 10 1 0 1 2 3 4 5 6 7 8 9
117

RESULTS

CLUSTERF-AI 200 blocks - 34 features

Clusters Overview & Evaluation

Point of Interest/ Block Akalyptos Shape Akalyptos Entrances No. Trees/Akalyptos No. Pools/Block

Point of Interest/ 200m neighborhood No. Parking/block Akalyptos Size Ratio built to surface block

S-ΧL (>300 sq) 0-3 diverse 62-83% trees: 2-11 pools: 0 parking: 0-4 POI: 0-7 POI: 25-167

XS-M (0-900 sq) 0-2 diverse 76-93% trees: 0-6 pools: 0-2 parking: 0-1 POI: 0-6 POI: 25-144

XS-L (0-1200sq) 0-2 diverse 74-94% trees: 0-6 pools: 0 parking: 0-2 POI: 0-18 POI: 40-260

S-XL (>300 sq) 0-2 diverse 73-85% trees: 0-6 pools: 0 parking: 0-3 POI: 11-24 POI: 88-279

XS-S, XL

(0-600, >1200sq), 0-1 diverse 71-90% trees: 0-6 pools: 0-1 parking: 0 POI: 0-3 POI: 26-83

XS-L (300-1200sq) 0-1 diverse 80-94% trees: 0-6 pools: 0-2 parking: 0-3 POI: 0-5 POI: 33-137

S-XL (>300) 0-2 diverse 65-90% trees: 0-8 pools: 0-2 parking: 0 POI: 4-21 POI: 105-325

L-XL (>900sq) 1-2 diverse 71-84% trees: 2-16 pools: 0 parking: 6-8 POI: 10-24 POI: 128-212

: XS (≤300sq) 0 polarized_center 99,96% trees: 0 pools: 0 parking: 0 POI: 1 POI:146

XS-M, XL

(0-900, >1200sq), 0-3 diverse 80-95% trees: 2-11 pools: 0-6 parking: 0-1 POI: 0-6 POI:36-168

0 1 2 3 4 5 6 7 8 9
118

Average height/ Block Average Speed/ Block Interest/ neighborhood

Transportation points/ 200m neighborhood Average Speed/ 200m neighborhood

Distance to Commercial Triangle Distance to Park Akalyptos Sunlighthours Summer Solstice 25-144 13.8-21.8m 21-50 26-56 0-14 0.45-2.1km 2.1-3.3km 0-13.4 1-3

25-167 10.42-20.33m 18-40 19.73-52.5 0-13 0.19-2.2km 1.0-3.3km 0-13.9 1-3 40-260 14-24.3m 10-43 10-40 0-23 0.3-1.5km 0.6-3.2km 0-12.8 1-2

Classification Medium area Akalyptos 88-279 9.7-19.7m 10-37 15-20 0-4 0.18-0.5km 0.7-1.0km 0-13.3 1-3 26-83 10.4-17.2m 10-50 18-45 0-11 0.04-1.5km 0.7-2.8km 7-14 1-2 33-137 12.3-21.5m 18-50 20-47 1-13 0.03-0.65km 0.9-2.5km 0-9.1 1-2 105-325 17.3-27.5m 17-64 : 18-39 3-27 0.25-1.5km 0.5-3.2km 0-13 1-3 128-212 17.3-27.5m 33-67 27-35 4-8 1.0-1.3km 2.9-3.2km 7.5-10.7 1-3

POI:146 20m 30 26 12 0.2km 0.9km 0 1

10.1-16.1m 0-43 17-32 3-10 0.06-0.35km 1.2-1.5km 0-7 1-2

POI:36-168
3 2 9 4 6 7 5 8 10 1 Order 119

RESULTS

CLUSTERF-AI 500 blocks - 19 features

After increasing our dataset to 500 blocks using however less input features we checked the correlation between them. As we see from the correlation coefficient heatmap underneath some parameters are very correlated such as akalyptos size type with the akalyptos area and the unpaved ratio with 90%.

Moreover the distance to park and the distance to commercial triangle are correlated with 72%, while the akalyptos closeness and the classification of the medium areas of the akalyptos are 70% correlated.

Last but not least the number of trees in the akalyptos is 58% correlated with the akalyptos size, the akalyptos size type and the unpaved ratio.

120

As a next step we wanted to define in how many clusters we should divide our dataset.

According to the graph Distortion Score Elbow for K-means Clustering we divided our dataset in 7 clusters.

We then exported our dataset as a csv in order to check the characteristics of each cluster. We gave the cluster types certain descriptions and we evaluated which clusters have better features.

121

RESULTS

CLUSTERF-AI 500 blocks - 19 features

Clusters Description

0: akalyptos size: XS-M (0-900 sq), aka_closeness: 0-2, akalyptos shape: diverse, park distance: 0.03-1.2km, aver. height buildings block: 10.14-27.54m, commercial triangle distance: 0.45-2.7km, trees: 4-11, pools: 0-4, public poi block: 0-2, num parking: 0-4, ratio_built_totalSurfBlock_type: 70-100%, classification_areas_m: 1-2, sunlighthours_longday: 0-11.66, longitudinal_Axis_orient: 9-81 degrees.

1: akalyptos size: L-XL (>900 sq), aka_closeness:0-3, akalyptos shape: diverse, park distance: 0.14-2.2km, aver. height buildings block: 10.42-21.4m, commercial triangle distance: 0.35-3.3km, trees: 1-16, pools: 0-6, public poi block: 0-2, num parking: 0-8, ratio_built_totalSurfBlock_type: 50-90%, classification_areas_m: 1-3, sunlighthours_longday: 0-14, longitudinal_Axis_orient: 0-45 degrees.

2: akalyptos size: XS (≤300 sq), aka_closeness:0, akalyptos shape: polarized center, park distance: 0.21km, aver. height buildings block: 20m, commercial triangle distance: 0.45km, trees: 0, pools: 0, public poi block: 0, num parking: 0, ratio_built_totalSurfBlock_type: 95-100% (99.958141%), classification_areas_m: 1, sunlighthours_longday: 0, longitudinal_Axis_orient: 9-18 degrees.

3: akalyptos size: XS-L (0-1200 sq), aka_closeness:0-1, akalyptos shape: diverse, park distance: 0.043-1.4km, aver. height buildings block: 9.73-24.33m, commercial triangle distance: 0.45-2.6km, trees: 0-6, pools: 0-1, public poi block: 0-1, num parking: 0-15, ratio_built_totalSurfBlock_type: 75-100%, classification_areas_m: 1-3, sunlighthours_longday: 6-14, longitudinal_Axis_orient: 0-36 degrees.

122

4: akalyptos size: XS-L (0-1200 sq), aka_closeness:0-4, akalyptos shape: diverse, park distance: 0.03-1.3km, aver. height buildings block: 10.43-23.75m, commercial triangle distance: 0.35-2.6km, trees: 0-8, pools: 0-2, public poi block: 0-1, num parking: 0-10, ratio_built_totalSurfBlock_type: 50-90%, classification_areas_m: 1-3, sunlighthours_longday: 0-14, longitudinal_Axis_orient: 27-63 degrees.

5: akalyptos size: S-XL (>300 sq), aka_closeness:1-3, akalyptos shape: diverse, park distance: 0.23-0.7km, aver. height buildings block: 12.38-18.10m, commercial triangle distance: 0.60-1.2km, trees: 0-4, pools: 0, public poi block: 0-1, num parking: 15-30, ratio_built_totalSurfBlock_type: 50-80%, classification_areas_m: 2-3, sunlighthours_longday: 0-14, longitudinal_Axis_orient: 9-54 degrees.

6: akalyptos size: XS-L (0-1200 sq), aka_closeness:0-2, akalyptos shape: diverse, park distance: 0.35-2.2km, aver. height buildings block: 10.41-21.81m, commercial triangle distance: 1.0-3.3km, trees: 0-8, pools: 0-2, public poi block: 0-1, num parking: 0-6, ratio_built_totalSurfBlock_type: 70-95%, classification_areas_m: 1-3, sunlighthours_longday: 0-14, longitudinal_Axis_orient: 0-54 degrees.

123

RESULTS

CLUSTERF-AI 500 blocks - 19 features

Clusters Overview & Evaluation

Akalyptos Shape Akalyptos Entrances No. Trees/Akalyptos No. Pools/Block No. Parking/block Akalyptos Size

Ratio built to surface block

XS-M (0-900 sq) 0-2 diverse 70-100% trees: 4-11 pools: 0-4 parking: 0-4

L-XL (>900 sq) 0-3 diverse 55-90% trees: 1-16 pools: 0-6 parking: 0-8

XS (≤300 sq) 0 polarized center 95-100% trees: 0 pools: 0 parking: 0 (99.958141%)

XS-L (0-1200 sq) 0-1 diverse 75-100% trees: 0-6 pools: 0-1 parking: 0-15

XS-L (0-1200 sq) 0-4 diverse 50-90% trees: 0-8 pools: 0-2 parking: 0-10

S-XL (>300 sq) 1-3 diverse 50-80% trees: 0-4 pools: 0 parking: 15-30

XS-L (0-1200 sq) 0-2 diverse 70-95% trees: 0-8 pools: 0-2 parking: 0-6

0 1 2 3 4 5 6
124

Public Point of Interest/ Block Average height/ Block Distance to Commercial Triangle Distance to Park Akalyptos Sunlighthours Summer Solstice

Classification Medium area Akalyptos

POI: 0-2 10.14-27.54m 0.03-1.2km 0.45-2.7km 0-11.66 1-2

POI: 0-2 10.42-21.4m 0.14-2.2km 0.35-3.3km 0-14 1-3

POI: 0 20m 0.21km 0.45km 0 1

POI: 0-1 9.73-24.33m 0.043-1.4km 0.45-2.6km 6-14 1-3

POI: 0-1 10.4-23.75m 0.03-1.3km 0.35-2.6km 0-14 1-3

POI: 0-1 12.38-18.10m 0.23-0.7km 0.60-1.2km 0-14 2-3

POI: 0-1 10.41-21.81m 0.35-2.2km 1.0-3.3km 0-14 1-3

2 1 7 6 3 5

4 Order

Akalyptos with very good potentials if the parking lots will be reduced 125

RESULTS

CLUSTERF-AI 500 blocks - 19 features

NUMBER CLUSTER EVALUATION DESCRIPTION

A lot of trees and pools, quite close to the park and commercial triangle, quite a lot of sunlight hours

Environmentally best performing one, many trees and water features (pools), ideal for all uses

TOTAL SCORE

Worse akalyptos. Ratio built/ total surface block (99.96%), no trees, no water features, 0 sunlight hours

Many of parking places, very few or no water features, ratio built/ surface block high, however sufficient sunlight hours

Very close to park, buildings height low, sufficient number of trees and some water features, many entrances

The Ratio built/ total surface block is very good, with sufficient trees and some water features. However, many parking lots, due to the commercial use as parking spaces in the uncovered area of the block. Ideal for neighborhood and public use. The Akalyptos cluster has the most potential to be turned into an oasis in the city center if the parking lots are replaced with elements that improve the microclimate.

A Sufficient number of trees, some water features, parking lots could be reduced and replaced with elements that Improve the microclimate

7 6 3 5 4 0 1 2 3 4 5 6 B A G F C E D 126
2 1

Image Credits: https://www.smarksthespots.com/numismatic-museum-cafe-athens/

Image Credits: https://www.sgd.org.uk/garden_design_journal/features/153/project_city_oasis_garden/

127
RESULTS WEBSITE SHOWCASE / CLUSTERF-AI 500 blocks AKALYPTOS CLUSTERING CLUSTER 0 CLUSTER 1 CLUSTER 2 CLUSTER 3 CLUSTER 4 CLUSTER 5 CLUSTER 6
128
Strefi Hill
Num_parking: 0
● Public_poi_block: 0
3 129
● Ratio_built/totalSurfBlock: 83.71% ● Sun_hours_longestDay: 9.87 h ● Deviation_North: 27-36° counterclockwise CLUSTER:

RESULTS

CLASSIFICATION MODELS

Learning Model
Logistic Regression machine learning Classification blocks
Performing SL model
Learning
Performing ANN model
Shallow
Forest Classifier XGBoost
Best
Deep
Model Arch_3200 epochs Arch_2200 epochs Arch_1200 epochs Best
Environmental Social Private Usage of Social Subarea Residential Neighborhood Public Public Neighborhood Residents 130
Classification Method in depth

Classification Methodology

The other machine learning model we applied was classification, a supervised machine learning technique. In this project, we wanted to take the medium sub-areas that were generated in grasshopper to another level of organizing, so we decided to classify the medium sub-areas inside the akalyptos for social inclusion, whether it is for the inhabitants of the block, the neighborhood usage, or the public. To compare the results correctly, we run a clustering model o the 200 blocks and another on the 500 blocks and evaluate which performed better. For the classification: we run three shallow learning models: Logistic regression, XGboost, and Random first classifier, and six deep learning models ANN with different layers architecture, and since we have to do that process on the large dataset (500 blocks) and the small one (200 blocks), we ended up running 12 classification models in total.

Classification Methodology 3 SL models for each patch 3 ANN models for each patch 6 models were trained for the 200 patch, and 6 models for the 500 patch In total we trained 12 classification models, and chose the best performing one. Best Performing model 131
RESULTS CLASSIF-AI 200 blocks - 34 features Shallow Learning Logistic Regression Xboost Random Forest Classifier Model Score 0.9666666666666667 Classification Report precision recall f1-score support 1_residents 1.00 0.98 0.99 43 2_neighbourhood 0.86 1.00 0.92 12 3_public 1.00 0.80 0.89 5 accuracy 0.97 60 macro avg 0.95 0.93 0.93 60 weighted avg 0.97 0.97 0.97 60 Cross Value Scores Conf_vals [1. 1. 0.96428571 1. 0.92857143] Confusion Matrix Values [[TP, FN, FP, TN] [42, 1, 0, 17], [12, 0, 2, 46], [4, 1, 0, 55] Model Score 1.0 Classification Report precision recall f1-score support 1_residents 1.00 1.00 1.00 43 2_neighbourhood 1.00 1.00 1.00 12 3_public 1.00 1.00 1.00 5 accuracy 1.00 60 macro avg 1.00 1.00 1.00 60 weighted avg 1.00 1.00 1.00 60 Cross Value Scores Conf_vals [1. 1. 1. 1. 1.] Confusion Matrix Values [[TP, FN, FP, TN] [43, 0, 0, 17], [12, 0, 0, 48], 5, 0, 0, 55] Model Score 0.95 Classification Report precision recall f1-score support 1_residents 1.00 1.00 1.00 43 2_neighbourhood 0.80 1.00 1.00 12 3_public 1.00 0.40 0.57 5 accuracy 0.95 60 macro avg 0.93 0.80 0.82 60 weighted avg 0.96 0.95 0.94 60 Cross Value Scores Conf_vals [1. 1. 0.96428571 0.96428571 0.96428571] Confusion Matrix Values [[TP, FN, FP, TN] [43, 0, 0, 17], [12, 0, 3, 45], [2, 3, 0, 55] Choosing the Best Performing SL Model- 200 Blocks For the 200 blocks we tested three shallow learning models: Logistic Regression, XGBoost, and Random Forest Classifier. The XGboost performed better with score 1. 132
CLASSIF-AI 500 blocks - 19 features Logistic Regression Xboost Shallow Learning Random Forest Classifier Model Score 1.0 Classification Report precision recall f1-score support 1_residents 1.00 1.00 1.00 89 2_neighbourhood 1.00 1.00 1.00 50 3_public 1.00 1.00 1.00 11 accuracy 1.00 150 macro avg 1.00 1.00 1.00 150 weighted avg 1.00 1.00 1.00 150 Cross Value Scores Conf_vals [1. 1. 1. 1. 1.] Confusion Matrix Values [[TP, FN, FP, TN] [89, 0, 0, 61], [50, 0, 0, 100], [11, 0, 0, 139] Choosing the Best Performing SL Model - 500 Blocks For the 500 blocks we did the same as previous, and we noticed that all are giving a good performance with the same score as 1. Model Score 1.0 Classification Report precision recall f1-score support 1_residents 1.00 1.00 1.00 89 2_neighbourhood 1.00 1.00 1.00 50 3_public 1.00 1.00 1.00 11 accuracy 1.00 150 macro avg 1.00 1.00 1.00 150 weighted avg 1.00 1.00 1.00 150 Cross Value Scores Conf_vals [1. 1. 1. 1. 1.] Confusion Matrix Values [[TP, FN, FP, TN] [89, 0, 0, 61], [50, 0, 0, 100], [11, 0, 0, 139] Model Score 1.0 Classification Report precision recall f1-score support 1_residents 1.00 1.00 1.00 89 2_neighbourhood 1.00 1.00 1.00 50 3_public 1.00 1.00 1.00 11 accuracy 1.00 150 macro avg 1.00 1.00 1.00 150 weighted avg 1.00 1.00 1.00 150 Cross Value Scores Conf_vals [1. 1. 1. 1. 1.] Confusion Matrix Values [[TP, FN, FP, TN] [89, 0, 0, 61], [50, 0, 0, 100], [11, 0, 0, 139] 133

RESULTS

CLASSIF-AI 200 blocks - 34 features

Deep Learning

Architecture Model_01

200 epochs

32 batch Architecture Model_02 200 epochs

32 batch Architecture Model_03 200 epochs

32 batch Architecture Model_04 200 epochs

32 batch

Architecture Model_05 200 epochs

32 batch Architecture Model_06 200 epochs

32 batch

134

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dense(64, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 13ms/step - loss: 2.9394 - accuracy: 0.6500 Accuracy: 0.6499999761581421

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(32, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 6ms/step - loss: 3.0170 - accuracy: 0.6500 Accuracy: 0.6499999761581421

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 6ms/step - loss: 4.0678 - accuracy: 0.5750 Accuracy: 0.574999988079071

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(64, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 4ms/step - loss: 3.7853 - accuracy: 0.6500 Accuracy: 0.6499999761581421 model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(21, activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu')

10ms/step - loss: 3.6310 - accuracy: 0.7000 Accuracy: 0.699999988079071

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu')) model.add(tf.keras.layers.Dense(21, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

10ms/step - loss: 3.6310 - accuracy: 0.7000 Accuracy: 0.699999988079071

135

RESULTS

CLASSIF-AI 500 blocks - 19 features

Deep Learning

Architecture Model_01

200 epochs

32 batch Architecture Model_02 200 epochs

32 batch Architecture Model_03 200 epochs

32 batch Architecture Model_04 200 epochs

32 batch

Architecture Model_05 200 epochs

32 batch Architecture Model_06 200 epochs

32 batch

136

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dense(64, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 3ms/step - loss: 1.7382 - accuracy: 0.9000 Accuracy: 0.8999999761581421

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(32, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 4ms/step - loss: 1.4724 - accuracy: 0.8700 Accuracy: 0.8700000047683716

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 3ms/step - loss: 1.4488 - accuracy: 0.8700 Accuracy: 0.8700000047683716

Architecturmodel = tf.keras.models.Sequential() model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(64, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 3ms/step - loss: 1.7476 - accuracy: 0.9200 Accuracy: 0.9200000166893005 model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(21, activation='relu')) model.add(tf.keras.layers.Dropout(0.1)) model.add(tf.keras.layers.Dense(32, activation='relu')

0s 3ms/step - loss: 1.8136 - accuracy: 0.8700 Accuracy: 0.8700000047683716

model.add(tf.keras.layers.Dense(64, input_shape=(n_cols,), activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu')) model.add(tf.keras.layers.Dense(21, activation='relu')) model.add(tf.keras.layers.Dense(32, activation='relu'))

0s 3ms/step - loss: 1.5709 - accuracy: 0.9000

Accuracy: 0.8999999761581421

137
138
2 1 0 Public Neighborhood Residents Environmental Social Interaction Private 139

DEPLOYMENT

07
07 | 1 Evaluation 07 | 2 Website

DEPLOYMENT

EVALUATION

Thessaloniki

The goal of building our machine learning web application is to solve the city of Athens's problems, opening up the way to solve similar issues in other cities. The machine learning model can only be effective if we can test it first and check if it delivers an accurate value to the usage of the medium sub-areas (yellow ones). So we deployed it locally using Hops in Grasshopper on the 12 new blocks from the city of Thessaloniki. Before we dived into the deployment results, we extracted the values of the medium sub-areas of those 12 blocks; then, we compared both values to determine whether they matched. The result was positive. All 12 blocks accurately predicted the usage of the medium sub-areas.

Image Credits: https://www.discovergreece.com/macedonia/thessaloniki 142

12 blocks from the city of Thessaloniki.

Image Credits: https://www.discovergreece.com/macedonia/thessaloniki 143

Thessaloniki

DEPLOYMENT
social area is for PUBLIC 40°37'58.2"N 22°56'22.9"E 40°37'59.7"N 22°56'24.1"E social area is for RESIDENTS
2 1 0 Public Neighborhood Residents 2 1 0 Public Neighborhood Residents EVALUATION
th01
th02
144
th03 social area is for RESIDENTS
145
40°37'57.8"N 22°56'25.1"E th04 social area is for RESIDENTS 40°37'55.2"N 22°56'29.6"E 2 1 0 Public Neighborhood Residents 2 1 0 Public Neighborhood Residents

Thessaloniki

DEPLOYMENT
social area is for RESIDENTS social area is for RESIDENTS 2 1 0 Public Neighborhood Residents 2 1 0 Public Neighborhood Residents EVALUATION
th05 40°37'53.5"N 22°56'30.4"E
th06 40°37'54.9"N 22°56'31.9"E
146
for
2 1 0
2
147
th07 th08 40°37'51.9"N 22°56'33.4"E 40°37'53.0"N 22°56'34.7"E social area is for NEIGHBORHOOD social area is
RESIDENTS
Public Neighborhood Residents
1 0 Public Neighborhood Residents
DEPLOYMENT
th10 40°38'01.1"N 22°56'26.1"E 40°38'00.1"N 22°56'28.1"E social area is for RESIDENTS 2 1 0 Public Neighborhood Residents social area is for RESIDENTS 2 1 0 Public Neighborhood Residents EVALUATION
148
th09
Thessaloniki

social area is for RESIDENTS social area is for RESIDENTS

Public Neighborhood Residents

2 1 0

Public Neighborhood Residents

2 1 0 149

th11 th12
40°37'57.3"N 22°56'32.6"E 40°37'55.2"N 22°56'29.6"E

RESULTS

WEBSITE/ CLUSTERF-AI 500 blocks

150
151

RESULTS

WEBSITE/ CLASSIF-AI 500 blocks

152
153
1.00 2.00 3.00

RESULTS

WEBSITE/ VISUALISATION DATA MAP 500 blocks

RESULTS

WEBSITE/ VISUALISATION DATA MAP 500 blocks

157

DISCUSSION CONCLUSIONS

● PCA and quick model training were used to understand the different features added to the dataset and improve how to measure the confusing ones.

● For the clustering model, using more blocks with fewer features results in fewer clusters. For instance, for the 200 blocks with 34 features, we had to cluster them into ten categories, while for the dataset with 500 blocks and 19 features, we had to cluster them into seven categories.

● The main characteristics of the clusters in both models are the same, thus giving us very similar results.

● Both deep and shallow learning models for the classification with fewer features/ more data performed better.

● For the classification model, the shallow learning model performed way better than the deep learning one.

● For the SL of the 200 blocks /34 features, the XGboost performed best with a score of 1.0, leaving the Logistic Regression with a 0.966 score and the Random Forest Classifier with 0.

● While the SL for 500 blocks/19 feature, all the three models performed well with a score of 1.0

160

● The ANN Classification models had a worse performance with a range of accuracy of 0.57-0.70. The best performing algorithms with the 200 blocks/34 features was Arch_05, 200 epochs, and 32 batch, with an accuracy of 0.7000 and loss of 3.6310.

● The 500 blocks/19 features performed relatively better in ANN classification model Arch_04 with epochs, 32 batch, accuracy 0.9200, and loss 1.7476.

● In evaluating the Akalyptos of Thessaloniki, the model accurately predicted the usage of the medium sub-area.

161
credits: https://www.instagram.com/p/B5GZNAYJ5tU/
Photo

DISCUSSION

FURTHER STEPS

● For the dataset creation, we would continue adding more real data of Akalytos to have an efficient model that works for the whole administrative boundary of the city. From previous experience and explorations with a similar type of machine learning, the SOM or Kohonen interpolation had introduced unrealistic values, hindering the model’s performance.

● The documentation of the data trees was based on observations from Google Maps. It could have been more accurate to be on-site and check each Akalyptos.

● In the future, we would like to include a classification model for the roofs to check the area and hosted shape for solar panels or green rooftops and identify ideal usage of those spaces

● Some Akalyptos with higher elevations could be used for the Green facade to mitigate the heat waves in summer and add natural aesthetic values to the space.

● A possibility to connect the Akalyptos through pedestrian streets could activate the empty spaces in the blocks into connected green corridors.

● We would also like to run the wind analysis for every block and include this feature in our dataset.

● We want to train new models after removing the minor relevant features from the PCA to see if it improves the ANN models’ behaviors.

● Add more variety to features that are not well distributed, for example, the number of pools and parking in the block.

162

Connection of more urban blocks/ akalyptos through green street-networks

163

RESEARCH WALL

166
Akalyptos Project MiroBoard
167

BIBLIOGRAPHY

● 4022architects. (n.d.). 42_Akalyptos SKG. [online] Available at: https://www.4022architects.gr/akalyptos-skg [Accessed 13 Sep. 2022].

● Abdishakur (2019). Retrieving OpenStreetMap data in Python. [online] Medium. Available at: https://towardsdatascience.com/retrieving-openstreetmap-data-in-python-1777a4be45bb [Accessed 13 Sep. 2022].

● Adamopoulos, G. (2021). City of Athens Urban Data. [online] GitHub. Available at: https://github.com/GeorgeAdamon/city_of_athens_urban_data/blob/master/how-to/how-to-classify/images/st ep8_Succesful_Classification.JPG [Accessed 13 Sep. 2022].

● Αναγνώπουλος, Ν. and Anagnopoulos, N. (2020). Αστικές καλλιέργειες στο Βοτανικό. dspace.lib.ntua.gr. [online] doi:10.26240/heal.ntua.19358.

● AntonasA., Bisti, M., Bitnik, M., Bridle, J., Douvitsas, V., Dragona, D., Dragonas, P., Dwyre, C., Earth, D. and Harvey, A. (2018). TOMORROWS: Urban fictions for possible futures. Onassis Cultural Center - Athens.

● app.shadowmap.org. (n.d.). Shadowmap. [online] Available at: https://shadowmap.org/.

● ArchDaily. (2012). Venice Biennale 2012: Greek Pavilion. [online] Available at: https://www.archdaily.com/266398/venice-biennale-2012-greek-pavilion?ad_medium=gallery [Accessed 13 Sep. 2022].

● ArchDaily. (2013). Re-Think Athens Competition Entry / Gianmaria Socci Architecture + Also Known As Architects + Alkistis Thomidou. [online] Available at: https://www.archdaily.com/341010/re-think-athens-competition-entry-gianmaria-socci-architecture-also-known -as-architects-alkistis-thomidou?ad_medium=gallery [Accessed 13 Sep. 2022].

● ARCHETYPE. (n.d.). Η Μήτις στη μοντέρνα Αθήνα. [online] Available at: https://www.archetype.gr/blog/arthro/i-mitis-sti-monterna-athina#.W5F9cP15mCk [Accessed 13 Sep. 2022].

● ARCHETYPE. (n.d.). Η πολυκατοικία ως μυθολογία. [online] Available at: https://www.archetype.gr/blog/arthro/i-polikatikia-os-mithologia [Accessed 13 Sep. 2022].

● Athens Resilient City. (n.d.). Athens Resilient City - Αθήνα Ανθεκτική πόλη. [online] Available at: https://www.athens-resilientcity.gr/ [Accessed 13 Sep. 2022].

● AthensVoice. (n.d.). Η Αθήνα απέκτησε Σχέδιο Δράσης για το Κλίμα. [online] Available at: https://www.athensvoice.gr/epikairotita/ellada/762043/i-athina-ehei-strammeno-vlemma-sto-mellon-kai-klima/ [Accessed 13 Sep. 2022].

● Bakogiannis, E., Kyriakidis, C. and Milioni, T. (2019). The Struggle for Integration of Refugees and Immigrants: Examples of Integration of Refugees in European Cities and the Case Study of Athens. European Journal of Engineering and Formal Sciences, 3(1), p.18. doi:10.26417/ejef.v3i1.p18-27.

● Bielik, M. (n.d.). Magnetizing Floor Plan Generator | DeCodingSpaces Toolbox. [online] Available at: https://toolbox.decodingspaces.net/magnetizing-floor-plan-generator/.

168

● Bielik, M. (n.d.). Tutorial | 2D and 3D isovists for visibility analysis | DeCodingSpaces Toolbox. [online] Available at: https://toolbox.decodingspaces.net/tutorial-2d-and-3d-isovists-for-visibility-analysis/ [Accessed 13 Sep. 2022].

● blogs.worldbank.org. (n.d.). Quantifying public spaces for better quality of urban assets. [online] Available at: https://blogs.worldbank.org/sustainablecities/quantifying-public-spaces-better-quality-urban-assets.

● Chou, R. (2022). Tutorial 3 — Working with Spatial Data in Grasshopper. [online] Data Mining the City 2022. Available at: https://medium.com/data-mining-the-city-2022/tutorial-3-working-with-spatial-data-in-gh-a186ef0598ce [Accessed 13 Sep. 2022].

● City Issue. (n.d.). Projects. [online] Available at: https://cityissue.org/portfolio/ [Accessed 13 Sep. 2022].

● Genesis Lab. (n.d.). Genesis Lab. [online] Available at: https://genesis-lab.dev/ [Accessed 13 Sep. 2022].

● Geoff Boeing. (2016). OSMnx: Python for Street Networks. [online] Available at: https://geoffboeing.com/2016/11/osmnx-python-street-networks/.

● Giorgos, Α. από T. (n.d.). triantafyllou giorgos architect. [online] Available at: https://triantafylloug.blogspot.com/2020/09/richard-woditsch-he-public-private.html [Accessed 13 Sep. 2022].

● gis.epoleodomia.gov.gr. (n.d.). e-ΠΟΛΕΟΔΟΜΙΑ Γεωγραφική Αναζήτηση Πληροφοριών. [online] Available at: http://gis.epoleodomia.gov.gr/v11/#/23.7270/37.9840/19 [Accessed 13 Sep. 2022].

● GitHub. (2022). gis_tutorials. [online] Available at: https://github.com/CenterForSpatialResearch/gis_tutorials [Accessed 13 Sep. 2022].

● GmbH (https://www.klokantech.com/), K.T. (n.d.). Coordinates on a map - pick GPS lat & long or coordinates in a projection system. [online] epsg.io. Available at: https://epsg.io/map#srs=4326&x=23.732838&y=38.000978&z=19&layer=osm [Accessed 13 Sep. 2022].

● HuffPost Greece. (2020). Η σύγχρονη Αθήνα κι η πολυκατοικία της είναι ένα αμιγώς ελληνικό φαινόμενο. [online] Available at: https://www.huffingtonpost.gr/entry/e-seychrone-athena-ki-e-polekatoikia-tes-einai-ena-amiyos-elleniko-faino meno_gr_5f63529ac5b61845586780c3 [Accessed 13 Sep. 2022].

● IAAC Blog. (n.d.). FARMSCAPES // A Macro level approach for Urban Farming. [online] Available at: https://www.iaacblog.com/programs/farmscapes-macro-level-approach-urban-farming/ [Accessed 13 Sep. 2022].

● issuu.com. (n.d.). City Abstractions and Information Production Tools by George Adamopoulos - Issuu. [online] Available at: https://issuu.com/george_adamon/docs/presentation-minusfirstslide [Accessed 13 Sep. 2022].

● issuu.com. (n.d.). Made in Athens by Panos Dragonas - Issuu. [online] Available at: https://issuu.com/panosdragonas/docs/madeinathens_catalogue [Accessed 13 Sep. 2022].

169

● issuu.com. (n.d.). Re-STOA-ring Athens. Opportunities and Risks of the Athenian Stoa (MA-Thesis) by Anais Alfieri - Issuu. [online] Available at: https://issuu.com/anaisalfieri/docs/layout_20final_20draft._20publish_20online_202 [Accessed 13 Sep. 2022].

● Κατασκευές Κτιρίων. (2021). Re-ΣΤΟΑ-ring Athens: Ευκαιρίες και Κίνδυνοι της Αθηναϊκής Στοάς. [online] Available at: https://kataskevesktirion.gr/re- [Accessed 13 Sep. 2022].

● kinoussis.gr. (n.d.). KINOUSSIS INSPIRATION DESIGN GROUP - ONE ATHENS. [online] Available at: https://kinoussis.gr/one-athens.html [Accessed 13 Sep. 2022].

● Koenig, R. (n.d.). Urban Elements | DeCodingSpaces Toolbox. [online] Available at: https://toolbox.decodingspaces.net/urban-elements/.

● Micocci, F. (2016). Hidden Patterns Of Urban Transformations In Athens. Studio Magazine. [online] Available at: https://www.academia.edu/36421724/Hidden_Patterns_Of_Urban_Transformations_In_Athens?fbclid=IwAR3r e1mjThtKbfU9jYDkPcN3y-i1k_2--xO5KtIJLcH7LuoW5_U5pQ7OqQ8 [Accessed 13 Sep. 2022].

● Oikonomou, M. (2014). The transformation of the urban block in the European City. In: 21st International Seminar on Urban Form ‘OUR COMMON FUTURE IN URBAN MORPHOLOGY’

● Onassis.org. (2022a). [online] Available at: https://www.onassis.org/culture/publications/builders-housewives-and-the-construction-of-modern-athens [Accessed 13 Sep. 2022].

● Onassis.org. (2022b). [online] Available at: https://www.onassis.org/whats-on/re-think-athens-urban-challenges-2014-15/athenian-apartment-building-poly katoikia [Accessed 13 Sep. 2022].

● Rodrigo-Comino, J. and Ermini, B. (2021). Are Cities Truly Dispersed? A Long-Term Analysis of Vertical Profile of Settlements in Athens’ Metropolitan Region. Sustainability, 13(6), p.3365. doi:10.3390/su13063365.

● The Urban Transcripts Journal. (n.d.). Chorographia Atheniensis: the reconstruction of the pre-modern geography of Athens - Fabiano Micocci. [online] Available at: https://journal.urbantranscripts.org/article/chorographia-atheniensis-the-reconstruction-of-the-pre-modern-ge ography-of-athens-fabiano-micocci/ [Accessed 13 Sep. 2022].

● thesis.ekt.gr. (n.d.). Διατριβή: Δυνατότητες αναβάθμισης του ιστού των ελληνικών αστικών κέντρων: ο ρόλος του ενεργού οικοδομικού τετραγώνου - Κωδικός: 18894. [online] Available at: https://thesis.ekt.gr/thesisBookReader/id/18894#page/422/mode/2up [Accessed 13 Sep. 2022].

● Woditsch, R. and Geipel, F. (2015). PLURAL - Private and Public Spaces of the Polykatoikia in Athens. Berlin: Technische Universität Berlin.

● www.arcgis.com. (n.d.). Map Viewer. [online] Available at: https://www.arcgis.com/apps/mapviewer/index.html.

170

● www.arch.upatras.gr. (n.d.). 3 ATHENS STORIES: COLLECTIVE SPACES IN THE INACTIVE STRUCTURES OF POLYKATOIKIA | Architecture Department of Patras. [online] Available at: http://www.arch.upatras.gr/en/projects/archive/3-athens-stories-collective-spaces-in-the-inactive-structures-ofpolykatoikia-5351.html [Accessed 13 Sep. 2022].

● www.athensvoice.gr. (n.d.). Μπορεί η ‘πίσω αυλή’ της πολυκατοικίας σου να γίνει ‘πλατεία’ συνάντησης; [online] Available at: https://www.athensvoice.gr/politismos/design-arhitektoniki/315531/mporei-i-piso-ayli-tis-polykatoikias-soy-naginei/amp/ [Accessed 13 Sep. 2022].

● www.greekarchitects.gr. (n.d.). Articles - STUDENTS PROJECTS - DESIGN PROJECTS - PROJECTS2015Designing a city block. [online] Available at: https://www.greekarchitects.gr/en/projects2015/designing-a-city-block-id10865 [Accessed 13 Sep. 2022].

● www.lifo.gr. (2017a). 7 δράσεις που θα αλλάξουν την Αθήνα ως το 2030 | LiFO. [online] Available at: https://www.lifo.gr/now/athens/7-draseis-poy-tha-allaxoyn-tin-athina-os-2030 [Accessed 13 Sep. 2022].

● www.lifo.gr. (2017b). Πώς μπορεί να αλλάξει η Αθήνα με 6 προτάσεις μικρής κλίμακας | LiFO. [online] Available at: https://www.lifo.gr/now/athens/pos-mporei-na-allaxei-i-athina-me-6-protaseis-mikris-klimakas [Accessed 13 Sep. 2022].

● www.naftemporiki.gr. (2022). Κλίμα: Πώς η Αθήνα θα μειώσει τις εκπομπές αερίων του θερμοκηπίου κατά 61% έως το 2030. [online] Available at: https://www.naftemporiki.gr/story/1873474/klima-pos-i-athina-tha-meiosei-tis-ekpompes-aerion-tou-thermokipi ou-kata-61-eos-to-2030 [Accessed 13 Sep. 2022].

● www.opengov.gr. (n.d.). Άρθρο 17: Κατασκευές και φυτεύσεις στους ακάλυπτους χώρους και περιφράξεις | Yπουργείο Περιβάλλοντος και Ενέργειας. [online] Available at: http://www.opengov.gr/minenv/?p=3932 [Accessed 13 Sep. 2022].

● www.statistics.gr. (n.d.). Απογραφή Κτιρίων 2011 - ELSTAT. [online] Available at: https://www.statistics.gr/census-buildings-2011 [Accessed 13 Sep. 2022].

● www.youtube.com. (n.d.). Antiparochi – A Short Introduction. [online] Available at: https://www.youtube.com/watch?v=dvjFiopD9wA [Accessed 13 Sep. 2022].

● www.youtube.com. (n.d.). Climate change: Europe’s first heat officer | DW Documentary. [online] Available at: https://www.youtube.com/watch?v=nTq1Sd9N7E8&t=17s [Accessed 13 Sep. 2022].

171

AKALYPTOS ATHENS WEB APP explores the potential of urban blocks' "Akalyptos" spaces (uncovered areas). The Akalyptos Spaces could be a possible solution ensuring the survival of Athens from invisible but ever-present threats, global warming & urban heat island effect. Using AI Algorithms, we define those spaces' current score, their ideal usage & privacy character

It is a project of IAAC, developed in MaCAD 2021/22 by Jumana Hamdani & Maria Papadimitraki, Faculty: Angelos Chronis.

10
ACKNOWLEDGEMENT
172

We would like to thank our supervisor Angelos Chronis for his valuable guidance throughout our thesis project. We would also like to extend our appreciation to all the faculty members at the MaCaD program, especially the program director David Andrés León and our coordinator Laura Ruggeri for their unlimited support and the program's organization. Also, we would like to thank Hesham Shawky for his technical support throughout the year.

173
MaCAD 2022 - Master Thesis Project

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.