Lisbon´s Real Estate Analysis based on Proximity Calculations

i Mas e Deg ee P og am in Da a Science and Ad anced Analy ics Lisbon’s Real Es a e Analysis based on P oximi y Calcula ions Ma ia Madalena Jo ge Do Nascimen o Valé io P ojec Wo k submi ed o In e na ional Jou nal o Scien i ic and Resea ch Publica ions p esen ed as pa ial equi emen o ob aining he Mas e Deg ee P og am in Da a Science and Ad anced Analy ics NOVA In o ma ion Managemen School Ins i u o Supe io de Es a ís ica e Ges ão de In o mação Uni e sidade No a de Lisboa MDSAA i NOVA In o ma ion Managemen School Ins i u o Supe io de Es a ís ica e Ges ão de In o mação Uni e sidade No a de Lisboa LISBON’S REAL ESTATE ANLYSIS BASED ON PROXIMITY CALCULATIONS by Ma ia Madalena Jo ge do Nascimen o Valé io P ojec Wo k p esen ed as pa ial equi emen o ob aining he Mas e ’s deg ee in Ad anced Analy ics, wi h a Specializa ion in Business Analy ics Supe iso / Co Supe iso : Miguel Cas o Co Supe iso : B uno Ja dim ii 02 2023 STATEMENT OF INTEGRITY I he eby decla e ha ing conduc ed his academic wo k wi h in eg i y. I con i m ha I ha e no used plagia ism o any o m o undue use o in o ma ion o alsi ica ion o esul s along he p ocess leading o i s elabo a ion. I u he decla e ha I ha e ully acknowledge he Rules o Conduc and Code o Hono om he NOVA In o ma ion Managemen School. Lisbon, Feb ua y 2023 iii ABSTRACT The eal es a e ma ke is always changing and e ol ing and wi h sus ainabili y becoming an inc easingly impo an opic, he way he p ice o a p ope y is de e mined, and he ac o s aken in conside a ion, should e ol e as well. Inspi ed by he popula concep o Sma Ci ies, mo e speci ically he 15-minu e ci y app oach, which is a concep highly ocused on accessibili y and walkabili y in ci ies, di e en a iables we e calcula ed o assess each p ope y’s accessibili y and di e si y o ameni ies. Bo h hese ac o s a e di e en o each p ope y, depending on hei loca ion. This wo k p esen s an analysis o he eal es a e ma ke in Lisbon whe e, aside om he physical a ibu es o a habi a ion, he di e si y and accessibili y o di e ence se ices in each loca ion will be e alua ed and in eg a ed in he machine lea ning p ocess. The goal is o know he impac o each calcula ed measu e when p edic ing he p ice pe me e alue o each house, in o de o help unde s and why simila houses ac oss Lisbon ha e such dis inc i e p ices. Bo h he Euclidean Dis ance and he Ne wo k Dis ance we e used in he calcula ions. The dis ance o Tejo Ri e and he numbe o comme cial es ablishmen s wi hin a 15-minu e walk adius we e wo o he mos impo an ea u es in he p edic i e models es ed. Th ee di e en me hods we e es ed and imp o ed, elec ing he Random Fo es Reg esso as he bes he one and he one o be used in he inal model. The inal model had hal o he a iance in he a ge explained by he all he calcula ed ea u es, which makes his analysis a po en ial good ool o help ill he gap o a p edic i e model ha only ac o s in he physical cha ac e is ics o a house. KEYWORDS Lisbon; Sma Ci y; Real Es a e; Accessibili y; Mobili y Sus ainable De elopmen Goals (SGD): This Mas e ’s Thesis con ibu es o he Sus ainable De elopmen Goal o Sus ainable Ci ies and Communi ies. i INDEX 1. In oduc ion ................................................................................................................... 1 2. Li e a u e e iew ........................................................................................................... 3 2.1 The 15-Minu e Ci y App oach ............................................................................... 3 2.2 Real Es a e ............................................................................................................. 4 2.3 E alua ing Real es a e’s p ices h ough 15-minu e Ci y ea u es ......................... 5 3. Me hodology ................................................................................................................ 8 3.1 Dis ance Calcula ions ............................................................................................. 8 3.2 Me hods o unco e ela ionships ........................................................................ 9 3.2.1 Linea Reg ession ............................................................................................ 9 3.2.2 Decision T ee Reg esso ............................................................................... 10 3.2.4 Random Fo es Reg esso ............................................................................. 10 3.2.5 Lasso .………………………………………………………………………………………………………11 3.2.6 Spea man co ela ion ................................................................................... 11 3.3 E alua ion Me hods ............................................................................................ 11 3.3.1 R2 Sco e ......................................................................................................... 11 3.3.2 R2 Adjus ed Sco e .......................................................................................... 12 3.3.3 Mean Squa ed E o ..................................................................................... 12 3.3.4 Mean Absolu e Pe cen age E o ................................................................. 12 3.4 So wa e Used ..................................................................................................... 12 4. Resul s and discussion ................................................................................................ 14 4.1 Fea u e Analysis ................................................................................................... 14 4.2 Linea Reg ession Analysis ................................................................................... 16 4.3 Random Fo es Reg esso Analysis ...................................................................... 18 4.4 Resul s Compa ison ............................................................................................. 19 4.5 Discussion ............................................................................................................ 20 5. Conclusion .................................................................................................................. 24 6. Limi a ions and ecommenda ions o u u e wo ks ................................................. 25 Re e ences ....................................................................................................................... 26 Appendix.......................................................................................................................... 28 LIST OF FIGURES Figu e 1.1 - Lisbon’s pa ishes mos p essu ed by local accommoda ion ......................... 2 Figu e 2.1 - Dis ance co e ed by each way o commu ing ............................................... 4 Figu e 2.2 - F amewo k p oposed by Pozoukidou & Cha ziyiannaki ................................ 6 Figu e 2.3 - Pa is F amewo k Resul p esen ed by Pozoukidou & Cha ziyiannaki ........... 7 Figu e 3.1 - Euclidean dis ance and Ne wo k dis ance ..................................................... 8 Figu e 3.2 - Decision T ee Reg esso ............................................................................... 10 Figu e 3.3 - Random Fo es Reg esso ............................................................................ 11 Figu e 4.1 - Linea Reg ession be ween P ice/me e and Comme ce Coun ................. 17 Figu e 4.2 - Random Fo es Reg esso Fea u e Impo ance ........................................... 18 Figu e 4.3 - Final model's mos impo an ea u es ........................................................ 20 Figu e 4.5 - Linea Reg ession be ween P ice/me e and Wa e dis ance ..................... 21 Figu e 4.5 - Linea Reg ession be ween P ice/me e and Shannon Di e si y Index ....... 21 Figu e 4.6 - Rada cha o he ea u es in A enidas No as and San a Cla a ................. 21 i LIST OF TABLES Table 3.1 - Calcula ed Fea u es, Desc ip ion and Fo mula ............................................... 9 Table 4.1 - Lasso's Top 10 Impo ance ............................................................................ 14 Table 4.2 - Spea man's Co ela ion be ween some o he a iables .............................. 15 Table 4.3 - Linea Reg ession Coe icien s ...................................................................... 16 Table 4.4 - Linea Reg ession ea u es wi h VIF below en ............................................. 17 Table 4.5 - Linea Reg ession Model sco es .................................................................... 18 Table 4.6 - Random Fo es Reg esso sco es .................................................................. 19 Table 4.7 - Final model sco es ......................................................................................... 20 Table 4.8 - Random Fo es Reg esso sco es .................................................................. 23 ii LIST OF ABBREVIATIONS AND ACRONYMS MAPE Mean Absolu e Pe cen age E o . MSE Mean Squa ed E o . R2 R-Squa ed. SUSHI Sus ainable His o ic Ci y Dis ic s. VIF Va iance In la ion Fac o . 1 1. INTRODUCTION O e he yea s ci ies ha e become as a eas, e y o en wi h ca -o ien ed de elopmen s, cha ac e ized by a less appealing aes he ics o landscapes and an e e -g owing popula ion, so much so, ha i is expec ed ha hund eds o people will mig a e o ci ies in he nex decade (Lusche , 2021). Faced wi h his u u e scena io, he ecen heal h c isis o Co id-19 and he p olonga ed clima e b eakdown, mo e ci ies a e shi ing hei ocus o becoming mo e sus ainable, inclusi e and wi h quicke ways o commu ing (Pozoukidou & Cha ziyiannaki, 2021). These c ises ended up exposing he ue p oblems and agili ies wi hin ci ies and hei need o a esponse. One o he mos s iking ealiza ions was he ime sa ed by wo king emo ely Be o e he global Co id-19 pandemic people would spend an ex ensi e pe iod o ime commu ing o wo k e e y day, which no only is e y ime consuming bu also e y ha m ul o he en i onmen and public heal h (Johansson, 2017). This shi om he adi ional o ice wo k and igid wo kspaces o di e en wo k s yles, aking in conside a ion wha people enjoy and need o do (Taylo , 2021), has helped 15-minu e ci ies alloca e wo kspaces o speci ic a eas, allowing companies o ha e o he small o ices ac oss own whe e hei employees could go. No only i p omo es a be e li e-wo k balance, lea ing mo e ime o people o do wha b ings hem joy, while no commu ing o wo k e e y day, bu also i helps o educe emissions and diminishes he ans e abili y o COVID-19 (Mo eno, Allam, Chabaud, Gall, & P a long, 2021). Acco ding o an in e na ional coali ion o mayo s ocused on clima e change and sus ainabili y called C40 Ci ies, implemen ing he idea o a 15-minu e ci y could help he u ban a eas eco e om he inancial and economic de as a ion o he pandemic (C40 Ci ies Clima e Leade ship G oup, 2021). As he 15-minu e ci y concep g ows s onge , neighbo hoods wi hin ci ies will be ully ocused on ul illing needs like accessibili y, walkabili y, esidence densi y and land mix use, es o ing he u ban concep o p oximi y. Making he mos ou o he close loca ion be ween se ices, ac i i ies and enabling people o access di e en oppo uni ies wi hin hei u ban en i onmen (Sma Ci y, 2022). This p oximi y-based s a egies o e s people local access o a wide scope o ameni ies, like schools and p eschools, heal hca e se ices, social se ices, es au an s, en e ainmen and cul u al e en s, pa ks e c., e e y hing ha is c ucial o quali y o li e (Bouche , 2020). This ans o ma ion, howe e , migh be a double-edged swo d. As neighbo hoods p og ess in o small walkable a eas, he eal es a e alue o he p op ie ies wi hin hose neighbo hoods may e y likely inc ease, gi en he ac ha one o he mos impo an aspec s o a p op ie y is i s loca ion and i s p oximi y o poin s o in e es . Lisbon has become one o Eu ope’s sma es ci ies, ha ing ecei ed he Eu opean G een Capi al Awa d in 2020 (Sma Ci y Lisbon, 2018) and, o e he las decade, ampli ied he a ailable ways o commu ing in he ci y by building mo e bike pa hs, ea anging key loca ions like squa es, oundabou s and s ee s. One o he bigges pu poses o he s a egic plan o Lisbon o 2020 was o inc ease i s popula ion by p omo ing housing, aking sma -ci y ini ia i es ega ding daily li e and ageing (POR Lisboa 2014-2020). In 2019, be o e he Co id-19 pandemic, Po ugal had one o he mos dynamic eal es a e ma ke s wi hin he Wes e n Eu ope, mos ly due o i s ax incen i es g an ed o o eign buye s and hei gold isas (Almeida, 2019). In hese las ew yea s, Lisbon has become a ou ism magne wi h many o eign eal es a e in es o s ha eno a e p ope ies in o sho - e m lease p ope ies like Ai bnb (Wa en & Almeida, 2020). These ans o ma ions ha e caused an inc ease on Po ugal’s eal es a e p ices gi en ha now hey a e di ec ed o ou is s, who ha e g ea e pu chasing powe . Acco ding o he Eu os a , back in 2019, Po ugal had an inc ease o almos 10% in hei eal es a e p ices, he bigges wi hin Eu ope (Idealis a, 2021). As Lisbon and i s neighbo hoods con inue o change and become mo e sus ainable, inclusi e and wi h ameni ies close p oximi y, i is na u al o assume ha he eal es a e p ices in he capi al will ollow hese imp o emen s and, ul ima ely, inc ease. 8 3. METHODOLOGY As p e iously seen, he 15-minu e ci y concep has ou p inciples ha assess di e en aspec s o a ci y. This esea ch ocuses on he p inciple o p oximi y and i s impac on eal es a es, h ough a ious p edic i e models cons uc ed, based on calcula ed a iables like he numbe o ameni ies in a 15- minu e walk adius and bo h he Euclidean and Ne wo k dis ance be ween an ameni y and a p ope y, excluding all he in o ma ion ega ding he physical aspec s o he la e . These calcula ed a iables y o cha ac e ize and sco e each p ope y based on he 15-minu e ci y’s p inciple o p oximi y. This was he selec ed p inciple o he analysis gi en i highly ocuses on accessibili y and mobili y in a ci y. 3.1 DISTANCE CALCULATIONS Bo h he Euclidean dis ance and he Ne wo k dis ance we e used o calcula e he dis ance be ween ameni ies and each p ope y, bo h calcula ed in e y dis inc ways. The Euclidean dis ance calcula es he dis ance be ween wo poin s by using he Py hago as heo em (Equa ion 1), as o he Ne wo k Dis ance, i uses Lisbon’s pedes ian ne wo k o ind and calcula e he bes ou e om one poin o ano he , using nodes and a sea ch adius. The e is no speci ic o mula o ne wo k dis ance as i depends on he opology and con igu a ion o he ne wo k being used. To be mo e p ecise, he me hods behind hese dis ance alues a e he ollowing: d(x,y) = √∑ n i=1 (1) Figu e 3.1 - Euclidean dis ance (le ) Ne wo k dis ance ( igh ) Re ie ed om - h ps:// inyu l.com/medium-AxU-pla o m The sco es o each p ope y a e di ided in ype o ameni y, calcula ion me hod and scale, making i a o al o 47 ea u es. Fo example, he e is a ea u e ha calcula es he ne wo k dis ance o eme gency es ablishmen s and disca ds he 5 minu es o he ip, one ea u e ha calcula es he Euclidean dis ance o he wa e , which in his case is he Tejo Ri e , and many o he s. These calcula ed a iables con empla e se e al scena ios o y o cha ac e ize he loca ion o each p ope y as bes as possible. Table 1 con ains he de ails and calcula ion me hods o some o he ea u es. The ea u es ha a e missing a e calcula ed he same way and hey can be analyzed in Annex 1. 9 Name Desc ip ion Calcula ion Name Desc ip ion Calcula ion T anspo Coun Numbe o Public T anspo a ion op ion in a 15- minu e walk adius g oupby(['IMOVEL_ID', ' anspo _dis ']).agg( {'geome y': ' i s ', ' anspo _coun ' : 'coun '}) Comme ce Coun Numbe o comme cial es ablishmen s in a 15-minu e walk adius g oupby(['IMOVEL_ID', 'come ce_dis ']).agg( {'geome y':' i s ', 'come ce_coun ':'coun '}) Shannon Di e si y Index Measu es he di e si y o ameni ies in a 15-minu e walk adius 𝑯′=−∑𝒑𝒊 𝒍𝒏 𝒑𝒊 𝑺 𝒊=𝟏 S – N o he ameni ies pi – N o es ablishmen s by ype o ameni y A e age Ne wo k Dis ance A e age o he ime o all he di e en ameni ies wi hin a 15-minu e walk adius A g Ne wo k Dis ance = dis ances.g oupby( “index_o”, g oup_keys=False)['dis ance'].mean() T anspo Dis ance Euclidean dis ance o he closes public anspo a ion op ion ckd_ ee= cKDT ee(B) dis ,idx= ckd_ ee.que y(A, k = 1) idx = i emge e ( * idx) (B_ix) Wa e Dis ance Euclidean dis ance o he Tejo Ri e Wa e Dis ance = house_node.geome y .dis ance( i e _node.geome y) Go e nmen Sco e Sum o he dis ances, in minu es, o all go e nmen al buildings in a 15-minu e walk adius Sco e = sum([1/(1+dis ) o dis in a['dis ance']]) Go e nmen Sco e 2.5 Sum o he dis ances o go e nmen al buildings, disca ding 2.5 minu es Sco e_2.5 = sum([1/(1+dis +2.5) o dis in a['dis ance']]) Go e nmen Sco e 5 Sum o he dis ances o go e nmen al buildings, disca ding 5 minu es Sco e_5 = sum([1/(1+dis +5) o dis in a['dis ance']]) Go e nmen Sco e 10 Sum o he dis ances o go e nmen al buildings, disca ding 10 minu es Sco e_10 = sum([1/(1+dis +10) o dis in a['dis ance']]) Table 3.1 – The calcula ed ea u es, i s desc ip ion and o mula. 3.2 METHODS TO UNCOVER RELATIONSHIPS Wi h he emaining independen a iables, di e en ea u e impo ance echniques we e calcula ed o u he mo e imp o e he p edic i e powe o he inal model. Each echnique calcula es a sco e o he inpu a iables in ha gi en model. The me hods es ed in each model we e he Linea Reg ession, Decision T ee Reg esso , and he Random Fo es Reg esso . Lasso and Spea man co ela ion we e also analyzed. 3.2.1 Linea Reg ession The linea eg ession uses he alue o an independen a iable o p edic he alue o he dependen a iable, h ough a linea app oach. This analysis allows he use o assess he quali y o he a iables in p edic ing he ou come along wi h which o hese a iables a e mo e signi ican and how each o hem impac s he dependen a iable. In his phase o he esea ch he Linea model c ea ed o es he ea u e impo ance is simply an ins ance, wi h no pa ame e s associa ed. The linea eg ession o mula ep esen ed below (Equa ion 2). 10 Y = β0+β1X (2) Whe e: • Y – Dependen Va iable • 𝜷𝟎– Cons an / In e cep • 𝜷𝟏 – Slope / Coe icien • X – Value o he independen a iables A common phenomenon in a eg ession analysis is he Mul icollinea i y, i exis s when mul iple independen a iables a e co ela ed in a model. The Va iance In la ion Fac o (VIF) iden i ies mul icollinea i y by es ima ing how much he a iance o a eg ession coe icien is ampli ied due o mul icollinea i y in he model. VIF anges om 1 upwa ds, whe e 1 is conside ed no co ela ed and alues g ea e han 5 is conside ed highly co ela ed. The h eshold used in his case was o e ain he a iable wi h a VIF alue below o equal o 10. 3.2.2 Decision T ee Reg esso This me hod uses a s uc u e lowcha -like whe e all he possible anges o esul s a e con empla ed. In a decision ee he e’s nodes and b anches, which con ain he condi ion and he esul , espec i ely. Each obse a ion goes h ough his lowcha -like ee based on he alue o i s a iables and a model is ained o p edic u u e obse a ions. Simila ly o he Linea Reg ession, only an ins ance o he model was c ea ed o his pa . Figu e 3.2 - Decision T ee Reg esso . Re ie ed om h ps:// inyu l.com/geeks- o -geeks 3.3.3 Random Fo es Reg esso The Random Fo es Reg esso is an ensemble me hod composed o mul iple decision ees. The inal p edic ion o each obse a ion is he a e age esul o all he ou comes p edic ed by each ee. Gene ally, e u ns be e esul s han Decision T ee Reg esso me hod. Like he wo p e ious Reg ession models, only an ins ance o he Random Fo es Reg esso is c ea ed o check he ea u e impo ance based on his me hod. 11 Figu e 3.3 - Random Fo es Reg esso . Re ie ed om h ps:// inyu l.com/analy ics- idhya Besides his ea u e impo ance me hod wo o he ea u e selec ion echniques we e es ed, Lasso and he Spea man’s co ela ion. 3.3.4 Lasso The Leas Absolu e Sh inkage and Selec ion Ope a o is based on a eg ession analysis ha ocuses on enhancing he p edic ion accu acy o he models. This me hod is di ided in wo s eps, he egula iza ion o model pa ame e s by sh inking he eg ession coe icien s, and he ea u e selec ion, whe e e e y non-ze o alue is used in he p edic i e model. 3.3.5 Spea man co ela ion E en hough Spea man’s co ela ion is no exac ly a ea u e impo ance me hod, i is an impo an echnique ha measu es he s eng h o associa ion be ween wo anked ea u es by assessing hei mono onic ela ionships, whe he linea o no . Spea man’s co ela ion is, simply, he Pea son’s co ela ion be ween he ank alues o wo ea u es (Equa ion 3). ρ=1 − 6∑di2 n(n2−1) (3) ρ- Spea man’s ank co ela ion coe icien di- Di e ence be ween he wo anks o each obse a ion n – Numbe o Obse a ions 6 – Cons an Value 3.3 EVALUATION METHODS To accesses each model’s p edic i e powe , R2 Sco e, R2 Adjus ed Sco e, Mean Squa ed E o (MSE) and Mean Absolu e Pe cen age E o (MAPE) we e calcula ed. These sco es combined ansmi all he in o ma ion needed o unde s and i he selec ed a iables a e helping he models in p edic ing he a ge mo e accu a ely o no . 3.3.1 R2 Sco e R-Squa ed anges be ween 0 and 1 which indica es he amoun o a iance in he ou pu o he dependen a iable ha is p edic able om he independen a iable. A high alue would mean ha 12 mos o he changeabili y in he dependen a iable’s ou pu can be explained by he model. R-Squa ed indica es he p opo ion o poin s ha lie wi hin he line c ea ed by he eg ession equa ion (Equa ion 4), he e o e, a highe alue indica es be e esul s. R2= 1 − ∑(Yi−Yi)2 𝑛1=1 ∑(Yi−Y)2 𝑛1=1 (4) 3.3.2 R2 Adjus ed Sco e As we add new ea u es o a mul iple eg ession model, R-squa ed will ei he con inue o inc ease o s op, ega dless o hei quali y. R-squa ed Adjus ed elimina es his issue by aking in conside a ion he numbe o samples in he da ase , and he numbe o ea u es in he model, inc easing he sco e only when he newly added ea u es imp o e he model’s accu acy (Equa ion 5). Jus like he R- Squa ed, a highe alue indica ed be e esul s. Adjus ed R2= 1 − (1−𝑅2)(𝑁−1) 𝑁−𝑝−1 (5) 3.3.3 Mean Squa ed E o The MSE (Mean Squa ed E o ) measu es he o al e o in a model by calcula ing he a e age squa ed di e ence be ween he obse ed alues and he p edic ed ones. This calcula ion epea s i sel o all he obse a ions and a e wa d, all hose alues a e summed and di ided by he o al numbe o obse a ions (Equa ion 6). This me hod highligh s he la ge e o s, making i use ul when wo king wi h models whe e hese e o s mus be minimized. MSE = 1n ∑(Yi−Yi)2 n i=1 (6) In a eg ession, as he da a poin s ge close o he eg ession line, he e o in he model dec eases, p oducing mo e p ecise p edic ions. 3.3.4 Mean Absolu e Pe cen age E o Mean Absolu e Pe cen age E o , o MAPE, measu es he accu acy o a p edic i e model by calcula ing he a e age pe cen age e o s o he p edic ions. The e o is he di e ence be ween he ac ual and o ecas ed alue, so a smalle alue would mean a smalle MAPE and he e o e, be e esul s. 3.4 SOFTWARE USED To de elop his s udy, P ojec Jupy e and py hon lib a ies we e used. Jupy e no ebook was he selec ed ool o de elop and un he code o his analysis, due o i s as accessibili y o py hon lib a ies. The OSMnx (Boeing, 2022) is one o he py hon lib a ies ha was used, and i ca ied ou one o he mos impo an pa s o he p ocess which was collec ing all he a ailable da a ega ding all ypes o ameni ies (go e nmen , heal h, educa ion, inance, public anspo , leisu e, ood and i e ). “OSM” means “Open S ee Map” and i allows he use o download geospa ial da a, model and analyze eal-wo ld s ee ne wo ks. I allows use s o model walkable u ban ne wo ks wi h he use o py hon, whe e hese models can be isualized and analyzed. The GeoPandas lib a y was also e y help ul in o de o wo k wi h he coo dina es ha we e gi en, and use hem o calcula e he di e ence dis ances, and he e o e, hei sco es ega ding he a ious ameni ies. 13 The ollowing Gi Hub link con ains wo o he no ebooks c ea ed o calcula e he di e en ea u es used in his hesis: Gi Hub - Thesis eposi o y 14 4. RESULTS AND DISCUSSION In his ollowing chap e he esul s and indings o he analysis will be exposed and discussed, o a clea e in e p e a ion o he ou comes, his chap e will be o ganized in ou sub-chap e s, 1.Fea u e Analysis 2. Linea Reg ession Analysis, 3. Random Fo es Reg esso Analysis and 4.Discussion 4.1 FEATURE ANALYSIS Acco ding o he Lasso me hod, nine ea u es we e elimina ed and hi y- i e we e selec ed, mos o which also selec ed in he p e ious me hods. E en hough hi y- i e we e selec ed, only he bes wen y ea u es we e aken in o conside a ion. Some o he bes ea u es we e he “Shannon Di e si y Index” , “go e nmen sco e”, and “educa ion sco e 2.5”. Lasso also allows o unde s and he impac o each ea u e, o example he ea u e “Shannon Di e si y Index” has an impo ance alue o - 1423.94, which mean i is an impo an ea u e wi h an in e se co ela ion o he a ge . Some o he ea u es selec ed by Lasso a e ep esen ed below, he comple e able is p esen in Annex 2. Fea u e Impo ance Shannon Di e si y Index 1423.94 Go e nmen Sco e 1033.45 Educa ion Sco e 2.5 886.11 Go e nmen Sco e 2.5 847.97 Finance Sco e 2.5 526.85 Educa ion Sco e 10 433.22 Heal h Sco e 2.5 418.10 Finance Sco e 10 331.98 Leisu e Sco e 2.5 231.93 Food Sco e 207.02 Table 4.1 – Lasso’s Top 10 Impo ance The Spea man’s co ela ion also o e ed some insigh s, no as conc e e as he p e ious me hods bu ne e heless impo an . The mos in e es ing insigh was he co ela ion be ween he di e en sco es which indica e ha some es ablishmen s in luence he p esence o o he s. Fo example, he numbe o public anspo a ions wi hin a 15-minu e walk adius o a house is highly co ela ed wi h he ood es ablishmen s sco e nea ha same house, meaning ha in a eas whe e he e a e mo e es au an s, ma ke s e c., he public anspo a ion ne wo k o e s mo e op ions o a pe son o commu e. The same happens wi h he numbe o educa ional buildings, like high schools o uni e si ies, in a 15- minu e walk adius and he dis ance o heal h es ablishmen s, meaning ha schools o en ha e hea h and eme gency se ices nea by. Fea u es ha ha e a co ela ion highe han 0.8 should be disca ded, a leas one o hem, since his means ha bo h a iables con ey he same in o ma ion. Ha ing wo e y simila ea u es, ansmi ing he same in o ma ion, is no e y e ec i e. 15 The g aphic below shows he Spea man’s co ela ion be ween some o he a iables. 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 1.Comme ce Coun 1 0.7 0.6 0.9 0.8 -0.8 0.4 0.5 0.6 0.3 0.3 0.7 0.4 0.2 0.6 0.004 2.Public T anspo Sco e 0.7 1 0.7 0.7 0.6 -0.5 0.6 0.6 0.8 0.2 0.02 0.8 0.3 0.5 0.8 0.06 3.Finance Sco e 10 0.6 0.7 1 0.7 0.5 -0.4 0.6 0.7 0.9 0.2 0.1 0.9 0.4 0.4 0.8 -0.2 4.Public Building Coun 0.9 0.7 0.7 1 0.7 -0.6 0.5 0.6 0.7 0.4 0.3 0.7 0.5 0.3 0.6 -0.1 5.T anspo Coun 0.8 0.6 0.5 0.7 1 -0.5 0.3 0.4 0.5 0.3 0.03 0.6 0.4 0.1 0.5 - 0.007 6.Shannon Di e si y Index -0.8 -0.5 -0.4 -0.6 -0.5 1 -0.1 -0.4 -0.4 0.08 -0.3 -0.5 0.06 0.00 8 -0.3 - 0.008 7.Leisu e Sco e 2.5 0.4 0.6 0.6 0.5 0.3 -0.1 1 0.6 0.6 0.1 0.08 -0.3 -0.5 0.06 0.00 8 -0.04 8.En e ainm en Sco e 0.5 0.6 0.7 0.6 0.4 -0.4 0.6 1 0.7 - 0.03 0.09 0.8 0.4 0.1 0.5 -0.3 9.Go e nme n Sco e 5 0.6 0.8 0.9 0.7 0.5 -0.4 0.6 0.7 1 0.1 0.2 0.8 0.4 0.3 0.8 -0.2 10.Educa ion Coun 0.3 0.2 0.2 0.4 0.3 0.08 0.1 - 0.03 0.1 1 0.07 0.2 0.4 0.7 0.5 0.4 11.A e age Ne wo k Dis ance 0.3 0.02 0.1 0.3 0.03 -0.3 0.08 0.09 0.2 0.07 1 0.02 0.2 - 0.06 - 0.03 0.2 12.Food Sco e 2.5 0.7 0.8 0.9 0.7 0.6 -0.5 0.6 0.8 0.8 0.2 0.02 1 0.3 0.4 0.8 -0.2 13.G eenspa ces Coun 0.4 0.3 0.4 0.5 0.4 0.06 0.7 0.4 0.4 0.4 0.2 0.3 1 0.2 0.4 0 .03 14.Educa ion Sco e 10 0.2 0.5 0.4 0.3 0.1 0.00 8 0.3 0.1 0.3 0.7 - 0.06 0.4 0.2 1 0.7 0.4 15.Heal h Sco e 2.5 0.6 0.8 0.8 0.6 0.5 -0.3 0.6 0.5 0.8 0.5 - 0.03 0.8 0.4 0.7 1 0.1 16.Wa e Dis ance 0.004 0.06 -0.2 -0.1 - 0.007 - 0.00 8 -0.04 -0.3 -0.2 0.4 0.2 -0.2 0.03 0.4 0.1 1 Table 4.2 - Spea man’s Co ela ion be ween some o he a iables Each ameni y has a se o sco es, bu only one sco e o each se will be kep . Despi e he ac ha highly co ela ed ea u es should be emo ed, in his case and since each a iable e e s a di e en ameni y, some highly co ela ed ea u es will be kep . In he ollowing sec ions o he ea u e analysis echniques we e deployed in o de o be e unde s and hese calcula ed ea u es. 16 4.2 LINEAR REGRESSION ANALYSIS The Linea Reg ession ep esen a ion be ween a a ge a iable and an independen a iable is one o he bes ways o isualize hei co ela ion and impac . Thei coe icien s ansmi some o he mos impo an in o ma ion o unde s and hei beha io , he cons an and he slope. Table 4 con ains he linea eg ession o each o he ea u es selec ed o he inal model. Table 4.3 – Linea Reg ession coe icien s Wi h he alues o 𝛽0 and 𝛽1 i is easy o unde s and each a iable’s beha io . The slope indica es i he ea u e has a di ec o in e se o ela ionship h ough is sign, i he slope is nega i e hen a lowe alue o he ea u e in ques ion, will gene a e a highe Ta ge alue. Fo example, he ela ionship be ween he p ice pe me e and he numbe o comme ce es ablishmen s in a 15-minu e walk adius, “comme ce coun ”, has a endency o lowe p ices ha ing a lowe numbe o comme ce es ablishmen s nea by. These a iables ha e a posi i e linea ela ionship, and can be isualized below: Fea u e 𝜷𝟎 / In e cep 𝜷𝟏 / Slope Fea u e 𝜷𝟎 / In e cep 𝜷𝟏 / Slope En e ainmen Sco e 5 3426.65 246.07 Heal h Sco e 2.5 3610.76 65.60 Leisu e Sco e 2.5 3449.14 34.19 G eenspaces Coun 3520.37 3.28 Shannon Di e si y Index 5049.19 -1276.83 Public Buildings Coun 3274.14 19.84 Public T anspo Coun 3261.35 7.88 Public T anspo Sco e 3461.62 22.71 Educa ion Coun 3923.01 -8.44 Educa ion Sco e 10 3933.27 -35.19 Food Sco e 2.5 3571.87 12.46 Go e mmen Sco e 5 3394.80 188.82 Comme ce Coun 3376.03 0.63 Wa e Dis ance 4519.41 -0.72 Finance Sco e 10 3404.08 106.01 A e age Ne wo k Dis ance 3040.67 80.37 17 Figu e 4.1 - Linea Reg ession be ween P ice/me e and Comme ce Coun When accessing he Linea Reg ession’s ea u e impo ance esul s and excluding he a iables wi h a VIF (Va iance In la ion Fac o ) highe han en, se en ea u es emained which a e also conside ed impo an in he di e en ea u e impo ance models es ed. The selec ed a iables by he Linea Reg ession’s ea u e impo ance a e he ollowing: Fea u e VIF A e age Ne wo k Dis ance 1.7306 Wa e Dis ance 2.5302 Shannon Di e si y Index 4.3115 Educa ion Coun 5.0916 T anspo a ion Coun 5.1829 G eenspaces Coun 6.7085 Public Building Coun 8.3501 Table 4.4 – Linea Reg ession ea u es wi h a VIF below en The ea u es abo e a e inside he h eshold, and he e is no su p ise in he ou come since all o hese a iables a e ecu en ly labeled as impo an h oughou he di e en ea u e impo an analysis. Gi en ha he Linea Reg ession is one o he simples p edic i e me hods, i is na u al ha he esul s a e no op imal, howe e i is a e y good me hod o gain some insigh s ega ding each ea u e’s co ela ion o he a ge . 24 5. CONCLUSION O en, when app aising he alue o a p ope y, he le el o accessibili y and mobili y is no ac o ed in, he physical aspec s such as he a ea and he numbe o ooms, a e much mo e impo an when ying o p edic he p ice pe me e o a house, han he aspec s ega ding i s loca ion. Howe e , i had o be unde s ood he impac hese o he aspec s could ha e on he inal p ice. Inspi ed by he 15- minu e ci y concep and i s pilla o p oximi y, di e en a iables we e calcula ed in o de o unde s and he accessibili y a ound each p ope y. A o al o o y-se en ea u es we e c ea ed using Euclidean Dis ance, Ne wo k Dis ance, by coun ing he numbe o es ablishmen s by ameni y and a ew o he me hods. No all he ea u es should be ac o ed in he inal model hus, in o de o selec a se o a iables, di e en ea u e impo ance and ea u e selec ion me hods we e analyzed. Lasso, Spea man’s co ela ion and he ea u e impo ance o h ee eg ession me hods we e deployed and analyzed, lea ing only six een ea u es. These emaining ea u es we e used as inpu a iables in h ee di e en models, he Decision T ee Reg ession, he Linea Reg ession and he Random Fo es Reg esso . To e i y he quali y o each model ou di e en sco es we e calcula ed and, ou o he h ee ins ances c ea ed, he Random Fo es Reg esso was he one ha pe o med be e . All sco es had signi ican imp o emen s, specially he R2 sco e wi h a alue o 0.5, meaning ha hal o he a iance in he a ge a iable is explained by he selec ed ea u es. Wi h he inal model ob ained, a deepe analysis in o he mos impo an ea u es and hei impac on he a ge was made. Calcula ed ea u es such as “wa e dis ance”, “comme ce coun ”, “public anspo dis ance sco e” and “g een spaces dis ance sco e”, u ned ou o be qui e impo an o all he models, p o ing ha he accessibili y and loca ion o a house do ha e an impac on i s selling p ice. When app aise s say ha a p ope y’s loca ion is one o he mos impo an aspec s, i is ue. E en hough inding and unning he p edic i e model was no he main ocus, he sco es o he inal mode we e conside ably good, especially conside ing ha he inpu ed a iables we e idealized based on he concep o 15-minu e ci ies, om which di e en sco es, dis ances and coun s o ameni ies we e calcula ed. In conclusion, his analysis alone will no p edic he p ice pe me e o a house as accu a ely as wi h i s physical a ibu es ac o ed in bu , i will help explain he di e ence in p icing when wo houses wi h iden ical physical cha ac e is ics ha e highly dis inc i e p ices, as he e is a co ela ion be ween hese calcula ed a ibu es and he a ge a iable ha p o es his heo y. 25 6. LIMITATIONS AND RECOMMENDATIONS FOR FUTURE WORKS As s a ed p e iously, his analysis ocuses on unde s anding i he p oximi y p inciple o he 15-minu e ci y app oach has any impac on he p ice pe me e alue o a houses. The a iables ha suppo his analysis we e c ea ed based on wha he p inciple assess a he han wha migh be bes inpu s o he p edic ions. No only his bu he e may be mo e a iables o cha ac e ize a p ope y’s loca ion and accessibili y, ha we e no conside ed in his esea ch. The me hods used o he calcula ions o said a iables we e highly ime consuming, which made any al e a ion o co ec ion, ha o ced a e- calcula ion, a conside able se back. Besides he echnical limi a ions, he e we e also a ew opics ha we e no discussed in his hesis, and i can be seen as limi a ions as well. Despi e he con i ma ion ha he calcula ed ea u es impac he p edic i e models when es ima ing he p ice pe me e , he e is no analysis ha con i ms ha he mos expensi e a eas in Lisbon a e in ac he ones ha ha e a highe comme ce coun , o a lowe wa e dis ance. Ano he limi a ion is ha he impac o he calcula ed ea u es is only es ed o he ci y o Lisbon hus i canno used as a ac wen es ing i in o he ci ies. Las ly and p obably he bigges limi a ion is ha hese p edic ions and analysis is based on he alues o oday and his is e y ola ile ma ke , whe e he p ices can easily change and a y om ci y o ci y. Fo u u es wo ks, i would be ad isable o de elop a be e p edic i e model, a deepe analysis ega ding he machine lea ning pa o his p ocess, whe e a mo e complex p edic i e model, wi h he ideal pa ame e s, would educe he MAPE and MSE o ou inal model, as well as inc ease i s R-squa ed and R2 Adjus ed sco e. I may be in e es ing o es his analysis in di e en ci ies, whe e he 15-minu e concep is g owing, and he eal es a e alues migh me shi ing, in o de o solidi y his co ela ion be ween accessibili y ac o s and he alue o a p ope y. This analysis migh be a good ool o help ill he gap o a p edic i e model ha only ac o s in he physical cha ac e is ics o a p ope y. Rega ding he limi a ions p e iously p esen ed, he e is a good oppo uni y o u u e wo ks. A compa ison be ween he p ice pe me e o a houses, pe pa ish and he alues o he calcula ed ea u es would con i m i indeed he mos expensi e a eas ha e be e accessibili y ac o s o li e close o he i e . To e i y ha his impac no only occu s in p ope ies in he ci y o Lisbon, o he ci ies can also be analyzed, like Pa is, ha ollows he 15-minu e ci y concep , o o he ci ies ha don’ ha e hei accessibili y ac o s as well de eloped, and check i he same impac is obse ed o i i is di e en , and i i is ex emely di e en , why does i happen. 26 REFERENCES Allam, Z., Mo eno, C., Chabaud, D., & P a long, F. (2021). P oximi y-Based Planning and he "15- Minu e Ci y": A Sus ainable Model o he Ci y o he Fu u e. Almeida, H. (2019, Sep embe 19). Eu ope's Ho es P ope y Ma ke Is Ge ing Too Ho o Some. Re ie ed om Bloombe g: h ps://www.bloombe g.com/news/ ea u es/2019-09- 19/po ugal-is-eu ope-s-ho es -p ope y-ma ke - oo-ho - o -some And es Duany, R. S. (2021, Feb ua y 8). De ining he 15-minu e ci y. Re ie ed om CNU : h ps://www.cnu.o g/publicsqua e/2021/02/08/de ining-15-minu e-ci y Boeing, G. (2022). OSMnx 1.1.2. Re ie ed om OSMnx: h ps://osmnx. ead hedocs.io/en/s able/ Bo ges, I. M. (2021). Tou ism and an o ma ion dynamics in his o ical neighbo hoods: The case o Al ama. Bouche , D. (2020). Local Li ing, Rise o 20 Minu e Ci ies Pos -Co id. Re ie ed om The Deck: h ps:// hedeck.o g.au/queensland-s a ewide-all- egions/local-li ing- ise-o -20-minu e- ci ies-pos -co id/ C40 Ci ies Clima e Leade ship G oup, C. K. (2021, May). Why e e y ci y can bene i om a ‘15-minu e ci y’ ision. Re ie ed om C40 Knowledge Hub: h ps://www.c40knowledgehub.o g/s/a icle/Why-e e y-ci y-can-bene i - om-a-15-minu e- ci y- ision?language=en_US Cos a, M. D. (2018). O Mé odo de Me cado. In O Valo do Imobiliá io - P incípios, Fa o es e Técnicas de A aliação Imobiliá ia. Duany, A.; S eu e ille, R. (2021). De ining he 15-minu e ci y. Re ie ed om CNU: h ps://www.cnu.o g/publicsqua e/2021/02/08/de ining-15-minu e-ci y Gou eia, J. P. (2021, June). Ci y T ans o ma ion; Al ama - Lisboa (Po ugal). Re ie ed om Sus ainable His o ic Ci y Dis ic s. Idealis a. (2021, Janua y 21). P ope y p ices in Po ugal ha e soa ed in he las decade. Re ie ed om Idealis a News: h ps://www.idealis a.p /en/news/p ope y- o -sale-in- po ugal/2021/01/21/845-p ope y-p ices-in-po ugal-ha e-soa ed-in- he-las -decade Lusche , D. (2021, June 16). In oducing he 15- Minu e Ci y P ojec . Re ie ed om :15 Ci y: h ps://www.15minu eci y.com/blog/hello Mo eno, C., Allam, Z., Chabaud, D., Gall, C., & P a long, F. (2021). In oducing he "15-Minu e Ci y": Sus ainabili y, Resilience and Place Iden i y in Fu u e Pos -Pandemic Ci ies. Sma Ci ies. O'Sulli an, F. (2020) Pa is Mayo : I 's Time o a '15-Minu e Ci y'. Re ie ed om Bloombe g: h ps://www.bloombe g.com/news/a icles/2020-02-18/pa is-mayo -pledges-a-g eene -15- minu e-ci y Pisano, C. (2020). S a egies o Pos - COVID Ci ies: An Insigh o Pa is En Commun and Milano 2020. Sus ainabili y. POR Lisboa 2014-2020. (n.d.). Re ie ed om Lisboa Po ugal2020: h ps://lisboa.po ugal2020.p /np4/%7B$clien Se le Pa h%7D/?newsId=253& ileName=ISE _CH_FEDER_PI_9.7_9.8_10.5.pd Pozoukidou, G., & Cha ziyiannaki, Z. (2021). 15‐Minu e Ci y: Decomposing he New U ban Planning’s Eu opia. Sus ainabili y, 5 - 6. Sma Ci y Tech. (2022). HOW SMART CITIES COULD HELP US ACHIEVE EQUITY AND ACCESSIBILITY. Re ie ed om Sma Ci y P ess: h ps://sma ci y.p ess/sma -ci ies-equi y-and-accessibili y/ Sma Ci y Lisbon. (2018, Sep embe 03). Re ie ed om Sma Ci y: h ps://sma ci y.b ussels/news- 598-sma -ci y-lisbon 27 S uyk, T. (2021, Feb ua y). The Fac o s o a 'Good' Loca ion. Re ie ed om In es opedia: h ps://www.in es opedia.com/ inancial-edge/0410/ he-5- ac o s-o -a-good-loca ion.aspx Taylo , P. (2021, Feb ua y). Whe e Did Ou Commu e Time Go? Re ie ed om Paul Taylo : h ps://pauli aylo .com/2021/02/05/whe e-did-ou -commu e- ime-go/ The Hindu Business Line (15 Sep embe 2016). Re ie ed om: h ps://www. hehindubusinessline.com/economy/sma -ci ies-could- esul -in-social- inequali y-say-expe s/a icle9111629.ece TED (Di ec o ). (n.d.). The 15-minu e ci y | Ca los Mo eno [Mo ion Pic u e]. Re ie ed om h ps://www.you ube.com/wa ch? =TQ2 4sJVXAI Wa en, H.; Almeida, H. (2020). Ai bnb Hos s Resis Lisbon’s Plan o F ee Up Housing. Re ie ed om Bloombe g: h ps://www.bloombe g.com/g aphics/2020-ai bnb-sho -le - e o ms- lisbon/ 28 APPENDIX Name Desc ip ion Calcula ion Name Desc ip ion Calcula ion T anspo Coun Numbe o Public T anspo a ion op ion in a 15- minu e walk adius g oupby(['IMOVEL_ID', ' anspo _dis ']).agg( {'geome y': ' i s ', ' anspo _coun ' : 'coun '}) Comme ce Coun Numbe o comme cial es ablishmen s in a 15-minu e walk adius g oupby(['IMOVEL_ID', 'come ce_dis ']).agg( {'geome y':' i s ', 'come ce_coun ':'coun '}) Educa ion Coun Numbe o Educa ional es ablishmen s in a 15-minu e walk adius g oupby(['IMOVEL_ID', 'educa ion_dis ']).agg( {'geome y':' i s ', 'educa ion_coun ':'coun '}) Public Buildings Coun Numbe o Public T anspo a ion op ion in a 15- minu e walk adius g oupby(['IMOVEL_ID', 'PubBuilding_dis ']).agg( {'geome y':' i s ', 'PubBuilding_coun ':'coun '}) Eme gency Coun Numbe o Eme gency es ablishmen s in a 15-minu e walk adius g oupby(['IMOVEL_ID', 'eme gency_dis ']).agg( {'geome y':' i s ', 'eme gency_coun ':'coun '}) G eenspaces Coun Numbe o Public T anspo a ion op ion in a 15- minu e walk adius g oupby(['IMOVEL_ID', 'g eenspaces_dis ']).agg( {'geome y':' i s ', 'g eenspaces_coun ':'coun '}) Shannon Di e si y Index Measu es he di e si y o ameni ies in a 15-minu e walk adius S – N o he ameni ies pi – N o es ablishmen s by ype o ameni y A e age Ne wo k Dis ance A e age o he ime o all he di e en ameni ies wi hin a 15- minu e walk adius A g Ne wo k Dis ance = dis ances.g oupby( “index_o”, g oup_keys=False)['dis ance'].mean() T anspo Dis ance Euclidean dis ance o he closes public anspo a ion op ion ckd_ ee= cKDT ee(B) dis ,idx= ckd_ ee.que y(A, k = 1) idx = i emge e ( * idx) (B_ix) Go e nmen Sco e Sum o he dis ances, in minu es, o all go e nmen al buildings in a 15-minu e walk adius Sco e = sum([1/(1+dis ) o dis in a['dis ance']]) Educa ion Dis ance Euclidean dis ance o he closes educa ional es ablishmen Calcula ed he same way as he T anspo Dis ance. Go e nmen Sco e 2.5 Sum o he dis ances o go e nmen al buildings, disca ding 2.5 minu es Sco e_2.5 = sum([1/(1+dis +2.5) o dis in a['dis ance']]) Eme gency Dis ance Euclidean dis ance o he closes eme gency building Calcula ed he same way as he T anspo Dis ance. Go e nmen Sco e 5 Sum o he dis ances o go e nmen al buildings, disca ding 5 minu es Sco e_5 = sum([1/(1+dis +5) o dis in a['dis ance']]) G eenspaces Dis ance Euclidean dis ance o he Calcula ed he same way as he T anspo Dis ance. Go e nmen Sco e 10 Sum o he dis ances o go e nmen al buildings, Sco e_10 = sum([1/(1+dis +10) o dis in a['dis ance']]) 29 Name Desc ip ion Calcula ion Name Desc ip ion Calcula ion closes g een a ea disca ding 10 minu es Comme ce Dis ance Euclidean dis ance o he closes comme cial es ablishmen Calcula ed he same way as he T anspo Dis ance. Public T anspo Sco e Sum o he dis ances, in minu es, o all public anspo a ions in a 15-minu e walk adius Sco e = sum([1/(1+dis ) o dis in a['dis ance']]) Public Building Dis ance Euclidean dis ance o he closes public Building Calcula ed he same way as he T anspo Dis ance. Public T anspo Sco e 2.5 Sum o he dis ances o public anspo a ions, disca ding 2.5 minu es Sco e_2.5 = sum([1/(1+dis +2.5) o dis in a['dis ance']]) Food Sco e Sum o he dis ances, in minu es, o all es ablishmen s ha sell ood in a 15-minu e walk Sco e = sum([1/(1+dis ) o dis in a['dis ance']]) Public T anspo Sco e 5 Sum o he dis ances, in minu es o public anspo a ions, disca ding 5 minu es Sco e_5 = sum([1/(1+dis +5) o dis in a['dis ance']]) Food Sco e 2.5 Sum o he dis ances o es ablishmen s ha sell ood, disca ding 2.5 minu es Sco e_2.5 = sum([1/(1+dis +2.5) o dis in a['dis ance']]) Public T anspo Sco e 10 Sum o he dis ances, in minu es o public anspo a ions, disca ding 10 minu es Sco e_10 = sum([1/(1+dis +10) o dis in a['dis ance']]) Food Sco e 5 Sum o he dis ances o es ablishmen s ha sell ood, disca ding 5 minu es Sco e_5 = sum([1/(1+dis +5) o dis in a['dis ance']]) Finance Sco e Sum o he dis ances, in minu es, o all inancial es ablishmen s in a 15-minu e walk adius Sco e calcula ed he same way as he p e ious ameni ies Food Sco e 10 Sum o he dis ances o es ablishmen s ha sells ood, disca ding 10 minu es Sco e_10 = sum([1/(1+dis +10) o dis in a['dis ance']]) Finance Sco e 2.5 Sum o he dis ances o inancial es ablishmen s, disca ding 2.5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 2.5 Leisu e Sco e Sum o he dis ances, in minu es, o all leisu e es ablishmen s in a 15-minu e walk adius Sco e calcula ed he same way as he p e ious ameni ies Finance Sco e 5 Sum o he dis ances o inancial es ablishmen s, disca ding 5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 5 Leisu e Sco e 2.5 Sum o he dis ances o leisu e es ablishmen s, disca ding 2.5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 2.5 Finance Sco e 10 Sum o he dis ances o inancial es ablishmen s, disca ding 10 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 10 30 Annex 1 – Table wi h all he calcula ed ea u es, hei names, desc ip ion and o mula. Name Desc ip ion Calcula ion Name Desc ip ion Calcula ion Leisu e Sco e 5 Sum o he Dis ance o leisu e es ablishmen s, disca ding 5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 5 Educa ion Sco e Sum o he dis ances, in minu es, o all educa ional es ablishmen s in a 15-minu e walk adius Sco e calcula ed he same way as he p e ious ameni ies Leisu e Sco e 10 Sum o he dis ance o leisu e es ablishmen s, disca ding 10 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 10 Educa ion Sco e 2.5 Sum o he dis ances o educa ional es ablishmen s, disca ding 2.5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 2.5 Heal h Sco e Sum o he dis ances, in minu es, o all heal h es ablishmen s in a 15-minu e walk adius Sco e calcula ed he same way as he p e ious ameni ies Educa ion Sco e 5 Sum o he dis ances o educa ional es ablishmen s, disca ding 5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 5 Heal h Sco e 2.5 Sum o he dis ances o heal h es ablishmen s, disca ding 2.5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 2.5 Educa ion Sco e 10 Sum o he dis ances o educa ional es ablishmen s, disca ding 10 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 10 Heal h Sco e 5 Sum o he dis ances o heal h es ablishmen s, disca ding 5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 5 En e ainmen Sco e 5 Sum o he dis ances o en e ainmen es ablishmen s, disca ding 5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 5 Heal h Sco e 10 Sum o he dis ance o heal h es ablishmen s, disca ding 10 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 10 En e ainmen Sco e 10 Sum o he dis ance o en e ainmen es ablishmen s, disca ding 10 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 10 En e ainmen Sco e Sum o he dis ances, in minu es, o all en e ainmen es ablishmen s in a 15-minu e walk adius Sco e calcula ed he same way as he p e ious ameni ies Wa e Dis ance Euclidean dis ance o he Tejo Ri e Wa e Dis ance = house_node.geome y.dis ance( i e _node.geome y) En e ainmen Sco e 2.5 Sum o he dis ances o en e ainmen es ablishmen s, disca ding 2.5 minu es Calcula ed he same way as he p e ious ameni ies o Sco e 2.5 31 Annex 2 – Lasso’s selec ed ea u es. Fea u e Impo ance Shannon Di e si y Index 1423.94 Go e nmen Sco e 1033.45 Educa ion Sco e 2.5 886.11 Go e nmen Sco e 2.5 847.97 Finance Sco e 2.5 526.85 Educa ion Sco e 10 433.22 Heal h Sco e 2.5 418.10 Finance Sco e 10 331.98 Leisu e Sco e 2.5 231.93 Food Sco e 207.02 Heal h Sco e 10 183.32 Leisu e Sco e 131.63 Go e nmen Sco e 5 114.11 Food Sco e 2.5 111.49 Heal h Sco e 105.49 Leisu e Sco e 10 59.53 En e ainmen Sco e 5 52.59 Public T anspo Sco e 2.5 39.63 32 Annex 3- Rada cha o A enidas No as (blue) and San a Cla a ( ed) a iables alue. 17