Identification and Classification of Routine Locations Using Anonymized Mobile Communication Data

  Ci a ion: Fe ei a, G.; Al es, A.; Veloso, M.; Ben o, C. Iden i ica ion and Classi ica ion o Rou ine Loca ions Using Anonymized Mobile Communica ion Da a. ISPRS In . J. Geo-In . 2022,11, 228. h ps:// doi.o g/10.3390/ijgi11040228 Academic Edi o s: Luca Pappala do and Wol gang Kainz Recei ed: 31 Decembe 2021 Accep ed: 25 Ma ch 2022 Published: 29 Ma ch 2022 Publishe ’s No e: MDPI s ays neu al wi h ega d o ju isdic ional claims in published maps and ins i u ional a il- ia ions. Copy igh : © 2022 by he au ho s. Licensee MDPI, Basel, Swi ze land. This a icle is an open access a icle dis ibu ed unde he e ms and condi ions o he C ea i e Commons A ibu ion (CC BY) license (h ps:// c ea i ecommons.o g/licenses/by/ 4.0/). In e na ional Jou nal o Geo-In o ma ion A icle Iden i ica ion and Classi ica ion o Rou ine Loca ions Using Anonymized Mobile Communica ion Da a Gonçalo Fe ei a 1,* , Ana Al es 1,2 , Ma co Veloso 1,3 and Ca los Ben o 1 1Cen e o In o ma ics and Sys ems (CISUC), Uni e si y o Coimb a, 3030-290 Coimb a, Po ugal; [email p o ec ed] (A.A.); [email p o ec ed] (M.V.); [email p o ec ed] (C.B.) 2Ins i u o Supe io de Engenha ia de Coimb a (ISEC), Poly echnic Ins i u e o Coimb a, 3030-199 Coimb a, Po ugal 3 Escola Supe io de Tecnologia e Ges ão de Oli ei a do Hospi al (ESTGOH), Poly echnic Ins i u e o Coimb a, 3030-199 Coimb a, Po ugal *Co espondence: g e [email p o ec ed] Abs ac : Digi al loca ion aces a e a ele an sou ce o insigh s in o how ci izens expe ience hei ci ies. P e ious wo ks using call de ail eco ds (CDRs) end o ocus on modeling he spa ial and empo al pa e ns o human mobili y, no paying much a en ion o he seman ics o places, hus ailing o model and enhance he unde s anding o he mo i a ions behind people’s mobili y. In his pape , we applied a me hodology o iden i ying indi idual use s’ ou ine loca ions and p opose an app oach o a aching seman ic meaning o hese loca ions. Speci ically, we used ci cula sec o s ha co espond o cellula an ennas’ signal a eas. In hose a eas, we ound ha all con ained poin s o in e es (POIs), ex ac ed hei mos impo an a ibu es (opening hou s, check-ins, ca ego y) and inco po a ed hem in o he classi ica ion. We conduc ed expe imen s wi h eal-wo ld da a om Coimb a, Po ugal, and he ini ial expe imen al esul s demons a e he e ec i eness o he p oposed me hodology o in e ac i i ies in he use ’s ou ine a eas. Keywo ds: call de ail eco ds; clus e ing algo i hms; human mobili y; meaning ul places; mobile phone da a; poin s o in e es 1. In oduc ion Human mobili y has become a p ominen esea ch ield in ecen yea s. The e is a g owing need o unde s and how people mo e and use u ban space in hei daily ou ines. We know o a ac ha human ajec o ies a e cha ac e ized by a high deg ee o empo al and spa ial egula i y. Fo each indi idual, he e is equen ime-independen a el dis ance and a signi ican p obabili y o e u ning o a ew highly equen ed loca ions, wi h only mino de ia ions o isi new des ina ions [ 1 ]. These hidden pa e ns o mo ion ha e impo ance in applica ions such as u ban planning, a ic o ecas ing and he sp ead o biological and mobile i uses. Ubiqui ous compu ing has in ac unlocked he po en ial o de e mine he pe sonal mo emen s o he masses ha p e iously we e only modeled using household su eys and na ional o egional census. In oday’s socie y, no hing could be mo e ubiqui ous han mobile phones, each o which is a po en ial senso p o iding a cons an da a s eam. On his basis, specialized spa io- empo al da ase s such as GPS eco ds ha e shown eno mous po en ial as a knowl- edge base abou human mobili y pa e ns. In spi e o ha , due o he o e head o col- lec ing and analyzing such de ailed and high equency loca ion logs a a la ge scale, many esea che s ha e been u ged o explo e o he da a sou ces as po en ial p oxies o human mobili y. Wi h his in mind, we ake special in e es in call de ail eco ds. A call de ail eco d (CDR) is a o m o da a ha documen s how a use in e ac s in de ail wi h he cellula ne wo k. These eco ds con ain in o ma ion ields such as o igin/des ina ion owe ID, use ID, ime o s a and du a ion on se e al ypes o communica ion. Collec ing he ISPRS In . J. Geo-In . 2022,11, 228. h ps://doi.o g/10.3390/ijgi11040228 h ps://www.mdpi.com/jou nal/ijgi ISPRS In . J. Geo-In . 2022,11, 228 2 o 18 cellula owe ID ha he use is connec ed o and co esponding Ca esian coo dina es means ha an accu a e loca ion is no possible, only an app oxima ion. In he case o CDRs, eco ds do no main ain egula ime in e als, unlike GPS, only when a ne wo k o use e en occu s (e.g., a call is made) he posi ion o he owe used is known. This leads o da a ha is bo h spa ially spa se and empo ally i egula . The e a e some compelling easons o using call de ail eco ds (CDRs) e sus mo e accu a e loca ion sou ces. Compa ed o o he loca ion log da a ypes, he o e head o collec ion and analysis is in e io . Fu he mo e, no applica ion has o be unning, no addi ional ba e y li e is consumed and da a collec ion canno be u ned o by he use , i is gene a ed by he egula usage o a mobile communica ion de ice and s o ed by he mobile ne wo k p o ide . Po en ially e e y phone in one p o ide ’s ne wo k can be used as a da a sou ce, esul ing in eno mous amoun s o in o ma ion ega ding a subs an ial pe cen age o he popula ion ha can be used o esea ch pu poses. Despi e he signi ican bene i s o scale on his in o ma ion ype, he de ail in he loca ions eco ded is a challenge o esea ch e o s ocused on indi idual use analysis. Since we only ha e he posi ion o a cellula owe , whose ange o ac ion spans be ween a ew hund ed me e s o se e al kilome e s, i is e y di icul o accu a ely pinpoin a use ’s loca ion. This unce ain y in loca ion makes i e y challenging o iden i y and classi y each use ’s ou ine places. Al hough his ques ion o ou ine places is a mo e s abilized wi h de ailed in o ma ion ypes, such as GPS [ 2 , 3 ], we eel ha he e is s ill oom o imp o emen and inno a ion wo king wi h call de ail eco ds. The main objec i e o his pape is o p esen an app oach o a aching seman ic meaning o a use ’s ou ine loca ions, in e ing he mo i a ions behind day- o-day mobili y. The ocus is on classi ying ac i i ies (e.g., shopping, dining, ou doo ec ea ion) ou side o he no mal home–wo k commu e. This s udy uses call de ail eco ds om Po uguese ci izens. Da a we e p o ided o his esea ch by one o he la ges elecommunica ion se ice p o ide s in Po ugal and eco ds a e om a ou -mon h pe iod be ween July and Oc obe 2020 by use s who had a majo i y o mobile e en s in ou s udy a ea, he dis ic o Coimb a. Using se e al ypes o complemen a y eco ds, including hose cap u ed wi hou human in e en ion (ne wo k-d i en e en s), we hope o su pass some o he issues ela ed o he spa se and i egula ime in e als o e en s. Filling he gaps o mobili y aces eases ajec o y econs uc ion and, as such, acili a es in e ing people’s ou ine places. Addi ionally, by using poin s o in e es (POIs) da ase s o classi y ou ine places, we ga he a weal h o in o ma ion ha can be used by o he s udies and applica ions. Analyzing each use ’s habi s and as es c ea es aluable pa ame e s o ecommenda ion sys ems, adap ing ma ke ing campaigns and imp o ing he quali y o se ice by aking in o accoun clien s p o iling. This pape is s uc u ed as ollows: Sec ion 2comp ises he s a e o he a , whe e we look o e he cu en bes p ac ices and implemen ed me hods. Sec ion 3p o ides an ou line o he me hodology including he esea ch app oach, da a analysis echniques and desc ip ion o he used algo i hms. In Sec ion 4, we epo expe imen s conduc ed o es he beha io o he p oposed me hodology as well as desc ibe and discuss he ob ained esul s. Valida ion and e alua ion a e also p esen ed since g ound- u h da a we e made a ailable om he mobile ne wo k p o ide . Finally, Sec ion 5add esses he conclusions and u u e wo k. 2. S a e o he A Wi h he inc easing popula i y o pe sonal mobile de ices and loca ion-based appli- ca ions, la ge-scale ajec o ies o indi iduals a e being eco ded and accumula ed a a as e a e han e e . Thus, i is possible o unde s and human mobili y om a da a-d i en pe spec i e. As a consequence o his con inuously inc easing a ailabili y o da a, wo ks based on spa ial- empo al da a ha e ecei ed a lo o in e es , wi h a la ge spec um o me hods de eloped. ISPRS In . J. Geo-In . 2022,11, 228 3 o 18 2.1. Call De ail Reco ds Despi e he lis ed bene i s, he use o his ype o da a aises ques ions ega ding he alidi y o p e ious wo ks and he ob ained conclusions, seeing ha CDRs p o ide limi ed accu acy along he spa ial dimension. In ac , s udies such as ha in [ 4 ] we e conduc ed wi h he objec i e o p o ing ha iden i ying use s’ mos signi ican loca ions, such as home and wo k, is possible wi h a high deg ee o success. Ano he esea ch wo k, a ound he same heme, Re . [ 5 ] compa ed CDR-based indi idual ajec o ies wi h e e ence in o ma ion om GPS logs. They ound hese wo ypes o in o ma ion o ma ch wi h a good enough accu acy o ex ac ing he use ’s mo emen s. The analysis o CDRs has al eady e ealed he spa ial ecu ence and empo al pe- iodici y o he mo emen pa e ns o people, who show a s ong endency o e u n o p e iously isi ed loca ions [ 1 ]. This en ails a high p edic abili y po en ial o human mobili y. Simila ly, impo an places in ou li es (e.g., home and wo kplace) can be in- e ed om CDRs [ 6 , 7 ]. Rou ine o hobby- ela ed loca ions ha e also been de ec ed wi h success [ 8 – 10 ]. The use o CDR da a in a el and ou ism is exempli ied in [ 11 ] whe e ou ism anspo a ion demand in Shanghai is in e ed by mobile phone da a and a sys em o p opose new ou es is de eloped. O he examples o p e ious wo ks making use o CDR analyses include he de ec ion and modeling o agg ega e mobili y lows a la ge scales [ 12 ], he cha ac e iza ion o indi idual mo emen pa e ns [ 13 ], o he compu a ion o o igin–des ina ion ma ices in u ban a eas [14]. P ese ing p i acy and da a p o ec ion is a conce n wo king wi h wha can be consid- e ed sensi i e pe sonal in o ma ion. To ha e ec , he wo k p esen ed in [ 15 ] demons a ed ha , wi h ce ain p o ec ion echniques applied, he e-iden i ica ion o an indi idual ia equen ly isi ed loca ions, co-loca ion pai s o spa ial empo al da a poin s wi h a high p obabili y is no possible. P ope s o age and e ie al echniques we e also discussed in [16] . Some new a emp s a e now p oposing p i acy p ese ing me hods o ajec o y da a based on CDR in o ma ion, anging om building ecommenda ion sys ems locally on a use de ice [ 3 ] o a no el s ay- egion based anonymiza ion echnique ha ca e s o impo an loca ions o a use [17]. 2.2. Poin s o In e es A poin o in e es (POI) is an en i y o in e es wi h a well-de ined loca ion. Poin s o in e es can ange om amous landma ks (e.g., museums, chu ches, owe s), na u al a ac ions (e.g., bays, coas s, wa e alls) o commonplace spo s (e.g., co ee shops, a e ns) [18]. The spa ial in e ac ions and dis ibu ions o POIs e eal di e en u ban unc ions which can be associa ed wi h ac i i ies. Fo di e en ypes o ac i i ies (e.g., spo s, ea ing, shopping), people can usually go o speci ic a eas. Fo he same eason, many scien i ic s udies (e.g., [ 2 , 3 , 10 , 19 – 21 ]) ha e ocused on ex ac ing ea u es o a ea classi ica ion. POIs can be collec ed om a ious online sou ces and a e equen ly a ailable o ee h ough applica ion p og amming in e aces (APIs) and online map se ice p o ide s. Fo example, Qihang e al. [ 2 ] p oposed o ma ch he isi o a use o a speci ic egis e ed loca ion using POIs ex ac ed ia Fou squa e’s API. In con as , P oux e al. [ 3 ] mapped in o ma ion om ou di e en geog aphic da abases (HERE, Fou squa e, G and-Lyon, IGN) and nine di e en social and cul u al da abases (P edi HQ, In e na ional Show ime, E enb i e, Songkick, Alle en s.in, Mee up, Spo ada , 10 imes) o pe o m use p o iling by de ec ing signi ican places and hei seman ic meaning wi h ex e nal sou ces o in o ma ion. 2.3. Seman ic Disambigua ion o CDRs The p ocess o seman ic en ichmen and he disambigua ion o places has mos ly been seen wi h he use o GPS ajec o ies. S udies such as [ 2 , 3 ] cen e hei e o on enue check-ins, whe e he goal is o ma ch a use o a isi o a egis e ed loca ion (e.g., es au an , ho el, e c.). These wo ks pe ec ly showcase he main di icul y wi h ying o disambigua e s op loca ions wi hou use inpu . Using CDRs u he augmen s he issues, as he mobili y aces a e much less p ecise. Some au ho s classi y geog aphic a eas by ca ego ies and ISPRS In . J. Geo-In . 2022,11, 228 4 o 18 ma ch hese a eas wi h he use ’s ou ines, ins ead o ying o ma ch spa se loca ions wi h a speci ic enue. Fo example, in [19], a i ual g id wi h cells o 500 by 500 m was cons uc ed. Fixed- size cells a e a common app oach o he classi ica ion o geog aphical a eas [ 20 – 22 ]. Each g id cell was classi ied acco ding o ou main ca ego ies (ea ing, shopping, en e ainmen and ec ea ional) and ma ched wi h he use ’s CDR loca ions. To ob ain ha classi ica ion, he numbe o POIs associa ed wi h each ac i i y ca ego y was eco ded o each cell, c ea ing an ac i i y dis ibu ion map. Each cell ac i i y p opo ion was hen no malized o a alue be ween he 0 and 1 and he K-means clus e ing algo i hm was applied o c ea e ou dis inc g oups. The inal s ep was inding he mos p obable ac i i y ca ego y o each o he k clus e s wi h a p obabili y unc ion. No g ound u h da a and no alida ion we e ca ied ou , as he au ho s ixa ed in hei conclusion on he di icul y and p i acy conce ns in ob aining such da a. In [ 10 ], he me opoli an a ea o Milan was di ided in o egions by using densi y clus e ing o agg ega e he g oups o p oxima e cell owe s. In sequence, he POIs ob ained om ou squa e we e used o classi y hese a eas by he mos equen ype o POIs p esen . S udy a ea subdi isions we e classi ied by he op le el ca ego ies in he POI da ase (e.g., shop, ood, nigh li e). The main ocus o he wo k was o ca ego ize explo e s, hose who a e inclined o b eak ou o hei daily mobili y ou ine and explo e new places, and ind ci y a eas associa ed wi h his beha io . Resul s and conclusions we e based on explo a o y analysis, inding, o example, he a eas and ca ego ies mos ela ed o explo a ion and a emp ing o alida e wi h exis ing knowledge o he s udy a ea. Expe imen s we e also conduc ed wi h Vo onoi diag ams, oad segmen a ion laye s, anspo a ion analysis zones (TAZ) and adminis a i e laye s [ 23 ]. This wo k p ima ily uses he s a is ics o CDRs o classi y geog aphical a eas and POIs as a complemen . Fi s , based on CDR da a, hey calcula e he pa ame e s equi ed including weekday and weekend CDR densi y spikes, numbe o peak alues, he in ensi y o peak alues, and he dis ibu ion o peak alues; his allows he a el beha io s and public cogni ion o be unde s ood. Then, POI ca ego y densi y was used o complemen he analysis o CDR e en s in each geog aphical di ision and based on ha , he iden i ica ion esul s a e modi ied. Valida ion was made h ough exis ing knowledge o he s udy a ea and compa ing he esul s ob ained o ha o he known eali y. Compa ed o he s a e o he a , he main no el ies in oduced by ou wo k a e he ollowing: (i) he in e ence o mobili y ou ines ou side o ou is and explo a ion ac i i ies, using a da ase mo e ailo ed o his analysis (Facebook Places); (ii) he ci cula sec o app oach, o c ea e signal a eas ela i e o each an enna is, acco ding o ou esea ch, an inno a i e way o subdi ide space o classi ica ion; (iii) in con as o p e ious wo k, ou app oach elies on mul iple POI ea u es (e.g., ca ego y pe cen age, popula i y and opening hou s) o ma ch loca ion da a wi h geog aphical a ea classi ica ion. 3. Ma e ials This sec ion p esen s, discusses and analyzes he da a ob ained o ga he ed in his wo k. Explo a o y da a analysis, using s a is ical g aphics and o he da a isualiza ion me hods, was conduc ed o summa ize he CDRs’ main cha ac e is ics. 3.1. CDR Da ase Fo he de elopmen wo k ca ied ou in his pape , a da ase comp ising 35,676 SIMs om he egion o Coimb a, Po ugal, was used. Da a collec ion co esponded o a pe iod o 4 mon hs, om 1 July 2020 o 30 Oc obe 2020, o aling 41,371,218 unique e en s. This amoun o da a was la ge enough o expe imen wi h and apply se e al s a e-o - he-a echniques and ob ain ep esen a i e esul s while s ill being accep able and manageable in e ms o size. Da a en ies we e a mix be ween e en -d i en and ne wo k-d i en en ies, which means ha some e en s did no equi e use pa icipa ion being gene a ed pe iodically ISPRS In . J. Geo-In . 2022,11, 228 5 o 18 wi hou human in e en ion. Each en y in he da a has: a use iden i ica ion ield; he imes amp o he e en ; a unique iden i ie o he cellula an enna; i s co esponding loca ion coo dina es; he an enna’s ini ial and inal angle o ac ion in deg ees; as well as he es ima ed ange alue in me e s. Be o e being made a ailable o esea ch, he use ’s iden i ie s we e pseudonymized. This means phone numbe s we e enc yp ed wi h a hash unc ion. This unc ion emained unknown o us, he esea che s, p ese ing he anonymi y o use iden i y. Analyzing he ob ained Call De ail Reco ds wi h ega d o he e en s made by each use , we can asce ain he alues p esen in Table 1. Fo he 35,676 indi iduals, he e is an a e age o app oxima ely 1346 unique e en s eco ded wi h a s anda d de ia ion o app oxima ely 425. The numbe o use s wi h mo e han 2000 e en s is small wi h he maximum eco ded being 7288. Table 1. Da ase analysis on numbe o e en s pe use . Use Coun 35,676 Mean E en s 1346.32 S d 425.81 Min 1.000 25% 1109.00 50% 1351.00 75% 1561.00 Max 7288.00 Figu e 1 ep esen s a his og am o he e en s pe use . As can be seen om Figu e 1 , he e en s do no ollow a no mal dis ibu ion. In ac , hey ha e an app eciable posi i e skewness while peaking a a ound he 1300 e en s ma k. Figu e 1. His og am o he numbe o e en s pe use . The pe iod o July/Oc obe 2020 ep esen s, a he ime o w i ing, he bes possible chance o no mal pa e ns o popula ion mo emen om he da a ha can be made a ailable o us since his ep esen s a mo e elaxed pe iod o COVID-19 con inemen in Po ugal [ 24 ]. E en hough we know i will ha dly gi e us an accu a e indica ion o p e- pandemic mobili y, i is, howe e , he closes we ob ained since he da a collec ion o his p ojec began. In sea ch o a be e indica ion o mobili y in he da a, we conduc ed an analysis o unique cell owe s isi ed by each use . This would se e as a be e indica ion i his da a con ains po en ial o mobili y s udies, o is comp omised by he a ypical si ua ion li ed h oughou he yea o 2020. ISPRS In . J. Geo-In . 2022,11, 228 6 o 18 The in o ma ion p esen ed in Table 2and Figu e 2shows ha he a e age use has 25 di e en loca ions eco ded and he s anda d de ia ion is ela i ely high, which should be expec ed as he e is a big sp ead o alues. This should ce i y ha al hough ime- ame o his ype o s udy is no ideal, he da ase con ains a good numbe o isi s o each pe son. Table 2. Da ase analysis on unique loca ions pe use . Use Coun 35,674 Mean (Unique Loca ions) 25.309 S d 18.406 Min 1.000 25% 12.000 50% 24.000 75% 34.000 Max 224.000 Figu e 2. His og am o unique loca ions pe use . 3.2. POIs Da ase POIs can be ex ac ed ia a single API o ob ained and agg ega ed om se e al sou ces. In ou pa icula case, i made mo e sense o use a single sou ce since places a e classi ied in o a ious ca ego ies co e ing a a ie y o subca ego ies, and he e a e o e lapping p oblems in di e en da ase s, so i would be necessa y o econs uc and eclassi y he POI da a o join wo o mo e sou ces. Facebook is p obably he mos popula social ne wo king si e ha makes i easy o people and/o businesses o connec and sha e wi h amily, iends and clien s online. Facebook Places is an associa ed geoloca ion se ice buil in o Facebook ha is designed o help use s sha e hei a o i e spo s and disco e new ones. Use s can “check in” a a ious loca ions, om ci ies o small s o es. Addi ionally, use s a e gi en he abili y o c ea e a new POI i he one hey in end o ’check-in’ o e iew does no al eady possess a Facebook Page. Business owne s can claim and ce i y he pages c ea ed by a hi d pa y by ollowing a e i ica ion p ocess. The main bene i o his da a sou ce when compa ed o Fou squa e ([ 2 , 10 , 25 ]) is he wide each o he Facebook pla o m and as such he amoun o POIs is inc eased as expec ed. The e is, in ac , a ep esen a ion o ca ego ies ha a e no p esen in Fou squa e’s da abase, including o ganiza ions, socie ies, inance and heal hca e. These ca ego ies, al hough no as impo an o ou ism o leisu e, a e impo an o in e e e yday mobili y ISPRS In . J. Geo-In . 2022,11, 228 7 o 18 mo i a ions o he esiden popula ion. Fu he mo e, by obse ing bo h da ase s, i was pe cei ed ha a bigge pe cen age o Facebook POIs con ained in o ma ion on opening and closing hou s. Fu he mo e, a da ase had al eady been cons uc ed o he whole coun y in a p e ious wo k [ 26 ]. This allowed access o an ex ensi e o line da abase using he code p o ided in he a o emen ioned wo k. In o al, he da ase has 221,724 unique poin s sp ead o e hund eds o ca ego ies o di e en hie a chies. An exce p o he POIs da a can be seen in Table 3. Table 3. Sample o he Facebook Places POI able. Name Check-Ins Hou s La i ude Longi ude Res au an e A iz 425 [[8, 0], [9, 0]] 39.82468 −7.4915 AZULMIR 15 [[9, 19], [9, 12]] 40.43211 −8.72678 B-Cul u e 0 [[9, 19], [9, 13]] 41.45011 −8.33808 ... ... ... ... ... Ca ego y Ci y Top Ca ego y Po uguese Res au an Cas elo B anco Food and Be e age Wholesale and Supply S o e Mi a Shopping and Re ail Medical and Heal h Guima ães Medical and Heal h ... ... ... 3.3. Use Su ey To ca y ou alida ion on p edic ed use ac i i ies, a su ey was made by he elecom- munica ions se ice p o ide as he in o ma ion needed was no p esen and could no be in e ed by any da a ga he ed o da e. The su ey was di ec ed o he exis ing use pool o he o iginal CDRs da ase in o de o compa e he knowledge ob ained by ou me hods wi h eali y. Since i was a olun a y ques ionnai e, i mean ha no all use s pa icipa ed. F om he o al 35,676 use s, only 574, o app oxima ely 1.61%, olun a ily ga e hei answe s. This ques ionnai e included in o ma ion such as: he p o essional ac i i y o he clien , wo k schedule, i he clien has a second home, whe e hey spend he weekend, main in e es s/habi s, and exe cise equency. 4. P oposed App oach F om his s udy and he analysis o he s a e-o - he-a esea ch, we c ea ed an ini ial oad map o expe imen a ion and me hods. The wo k can be di ided in o sec ions wi h he inal goal being, wi h CDRs as inpu , o ou pu a de ailed able o ou ine a eas and hei classi ica ion. 4.1. CDR P e-P ocessing P e-p ocessing he da ase included e ie ing and in e ing addi ional da a columns (e.g., he day o he week, wo kday/weekend) om he exis ing ones. This was in ended o ease he de ec ion o spa io- empo al pa e ns in he eco ds. An in ege o he day o he week ( om 0 o Monday o 6 o Sunday) and a Boolean alue o he wo kday o weekend (0 being a wo kday) we e ob ained om he imes amp columns. Fu he mo e, we adop ed he ime segmen di ision ound in [ 22 ]. Fo each en y, aking he imes amp, we e i ied he co esponding in e al. One day is di ided in o eigh ime segmen s o cap u e he in aday a ia ions in ac i i y pa icipa ion: ea ly mo ning (3–6 a.m.); mo ning—peak hou (6–9 a.m.); mo ning—wo k (9 a.m.–12 p.m.); noon (12–2 p.m.); a e noon—wo k(2–5 p.m.); a e noon—peak hou (5–8 p.m.); nigh (8 p.m.–12 a.m.); and midnigh (12–3 a.m.) [22]. ISPRS In . J. Geo-In . 2022,11, 228 8 o 18 As seen by da a explo a ion in he CDR da a desc ip ion sec ion, he e we e some cases whe e use s had a lowe numbe o e en s han a e age—e en hose use s wi h less han one e en pe day. As expec ed, hese will add li le o no in o ma ion o ou esea ch pu pose, since we seek a highe numbe o e en s in o de o in e spa ial pa e ns. Thus, we c ea ed a simple unc ion ha , aking as inpu an e en h eshold, il e s ou all use s wi h a numbe o e en s below ha h eshold. Fo example, emo ing use s wi h less han one e en pe day esul ed in a educ ion o 0.12% o he da ase o 25,858 unique e en s. Ano he s ep was he de ec ion o cellula owe eselec ion in he middle o calls, o in e y quick succession, c ea ing impossible ajec o ies when aking in accoun he speed o mo emen . This is due o au oma ic ne wo k load balancing, a phenomenon o en called load sha ing [ 7 ]. Wi h his in mind, dis ances be ween he ne wo k owe s we e compu ed and, consequen ly, he a eling speeds o use s we e es ima ed in consecu i e eco ds. Fo he de ec ion o he load sha ing e ec , a speed-based me hod was implemen ed. A sequence is iden i ied i he owe swi ching speed exceeds a gi en h eshold. We se he alue a 200 km/h inspi ed by he wo k o Io an e al. [27]. A e hese ini ial s eps and be o e we could sea ch o ou ine ac i i y pa e ns, we needed an accu a e iden i ica ion o each use ’s home and wo kplace loca ions. These a e mos likely he places whe e people spend he majo i y o hei ime and ep esen a la ge po ion o hei mobile eco ds. Finding hese loca ions i s is impo an because i allows us o ocus ou a en ion on ele an eco ds o ou esea ch o habi s ou side o hese places. Thank ully his opic has been a subjec o many p io s udies and he e a e p o en me hods wi h good accu acy. 4.2. Home and Wo kplace De ec ion Mo i a ed by Vanhoo e al.’s wo k [ 6 ], a mixed app oach o ime il e ing and densi y- based clus e ing is p oposed. Fi s ly, we selec ed he empo al in e als o sea ch when someone is no likely o be ound in he places we wan o iden i y. In his case, he home ime in e al was de ined as he pe iod om 7 p.m. o 9 a.m. as pe [ 6 ]. Howe e , because hey did no y hei me hod o wo kplace de ec ion, we de ined wo king hou s as he pe iod om 9 a.m. o 5 p.m., a common schedule o 8 h o day wo ke s. Addi ionally, wo kplace CDRs we e cons ained o wo kdays. Gi en he s a e-o - he-a esea ch, we op ed o densi y-based spa ial clus e ing, o DBSCAN, as pe he wo ks o [ 7 , 8 ]. DBSCAN is s ill o his day conside ed a compe en algo i hm o g ouping CDRs and inding impo an a eas. I s ecu en appea ance h oughou he li e a u e suppo ed ou choice o use i ou me hodology. A e iden i ying and excluding homes and wo kplaces om he indi idual use ’s da a, we a e le wi h he emaining loca ions. F om hese, we hen need o unde s and which a e he mos ele an o he daily ou ine, i.e., he mos isi ed ones ha accoun o a subs an ial ime expendi u e. 4.3. O he Rou ine Loca ions The chosen me hod o de ec he home and wo kplace using DBSCAN could also be used o ind o he ou ine loca ions. Wi hou he ime es ic ions o home/wo kplace hou s and by keeping all he clus e s, a he han highligh ing he one wi h he mos e en s, i would be a good candida e solu ion. The issue ound wi h using his densi y- based clus e ing is ha we would lose addi ional p ecision in pinpoin ing he exac use posi ion. An enna loca ions al eady ha e g ea unce ain y when i comes o ma ching he use posi ion, and clus e s consis ing o se e al an ennas would subs an ially inc ease he challenge. Fo ou ine loca ions, we wan o e ain he maximum p ecision possible. The la ge he a ea, he mo e di icul i will be o ma ch a speci ic ac i i y. Inspi ed by he wo k o Quad i e al’s [ 10 ], which di ided use s’ loca ions in classes o impo ance wi h espec o he numbe o unique isi days, a simila app oach was used. The h ee classes a e: mos isi ed places (MVPs), loca ions mos equen ly isi ed by he use ; occasionally isi ed places (OVP), loca ions o in e es o he use , bu only isi ed ISPRS In . J. Geo-In . 2022,11, 228 9 o 18 occasionally; excep ionally isi ed places (EVP): non- ou ine places. To classi y places in hese classes, a ele ance me ic was calcula ed o each place in he use ’s eco ds. The ini ial ele ance o a loca ion l o a ce ain use u : R(l , u) was calcula ed by he numbe o unique days ha he use isi ed he loca ion d isi (l , u) o e hei o al numbe o ac i e days d o al(u) . As ou main goal is no only o de ec ou ine loca ions bu also o in e ac i i ies, he ele ance me ic was modi ied o accommoda e he need o a ime window and day ype. We sepa a ed use places by coo dina es, ime in e al and ype o day (wo kday/weekend). The inal me ic o calcula ing he ele ance o a loca ion, R(l m,d , u) , is ha p esen ed in Equa ion 1. Ins ead o coun ing he unique days ha he use isi ed loca ion l , we coun ed he unique days ha he use isi ed l in ime in e al m and ype o day d: R(l m,d,u) = d isi (l m,d,u) d o al(u)(1) We used he calcula ed me ic as inpu o a K-means clus e ing algo i hm, his ime wi h inpu alue k = 3 o ob ain he h ee dis inc g oups. Figu e 3, a 3D sca e plo , shows coo dina e poin s clus e ing by he ele ance me ic o one selec ed pe son in he da a. No e ha he Z axis ep esen s he ele ance me ic while he X and Y a e la i ude and longi ude, espec i ely. Pu ple colo coded poin , wi h he highes ele ance sco e a e MVPs, wi h o ange poin s being OVPs and blue poin s EVPs. Figu e 3. 3D sca e plo o he K-means clus e ing applied o use loca ions wi h K = 3. Explo a ion o holiday- ela ed ac i i ies (EVPs) do no en ail a signi ican pa e n in he da a o be conside ed and a e no analyzed u he . The idea is ha excluding home and wo k, we ind o he equen ly isi ed places including MVPs and OVPs, ha ha e signi ican impo ance o each use . 4.4. Geog aphic Regions Classi ica ion To p o ide be e insigh in o he mo i a ions behind he mobili y, a his poin , we op ed o subdi ide he s udy a ea and classi y he esul ing geog aphic egions wi h he mos likely ac i i y. This is an impo an s ep in o de o ob ain he use ’s classi ied ou ine loca ions. The selec ion o he egions is impo an as he size and shape can in luence he inal esul s. A sligh ly la ge o di e en ly shaped egion can encompass mo e POIs, skewing he ac i i y classi ica ion. We needed well-de ined egion bounda ies ha ep esen ed he sea ch a ea in o de o a unc ion o e u n all con ained POIs. Se e al app oaches we e conside ed, including ixed size ([ 19 – 22 ]) and dynamically sized [ 23 ]; howe e , we p oposed a new ype o egion o ac i i y classi ica ion using he an enna’s signal a ibu es. ISPRS In . J. Geo-In . 2022,11, 228 16 o 18 Fu u e Wo k Some possible imp o emen s we e ound by conduc ing he analysis o a ea’s ac i i y classi ica ion. Manually gi ing a weigh o POIs o ce ain ypes o inc ease hei impo ance depending on he ime o day, e.g., o es au an s a egula meal hou s, would possibly change he ac i i ies o be e mi o popula ion endencies. The same e ec could also be achie ed wi h a popula i y/check-in alue ha was hou dependen , bu as a as we know, no POI da ase con ains his in o ma ion. The e is s ill he ques ion o poin s missing om he used da ase , as hey migh no be egis e ed in he used da a p o ide . One possible solu ion would be he combina ion o se e al POI da ase s wi h he added di icul y o me ging comple ely di e en ca ego y hie a chies in o one. The elecommunica ions se ice p o ide da a cu en ly con ain cells om 2G o 4G; howe e , we do no di e en ia e be ween hese cells. We unde s and ha he signal a eas o an ennas o di e en echnologies o e lap. Howe e , in ou app oach, hese cells will ha e he same classi ica ion, and should no a ec he p edic ed use ou ines we would like o explo e a way o me ge o e lapping an ennas and hei eco ds. In he u u e, in addi ion o conside ing he equency o isi a ion, i s empo al dis ibu ion (e e y day, weekly, biweekly) could be a ac o o ake in o accoun ega ding he ype o ac i i y. The a eas c ea ed o classi ica ion, al hough close o he eali y o whe e he use migh be, s ill emain oo la ge o ha e a good pe cen age o ce ain y in e ms o use ac i i y. Newe in o ma ion sou ces ha ha e been discussed wi h he elecommunica ion se ice p o ide o u u e wo k ha e he po en ial o imp o e he use ’s loca ion e en u he . Doing so allows o a smalle sea ch a ea and gene ally mo e accu a e me hods. The a i al o 5G ne wo ks, wi h mo e p ecise smalle adius an ennas, could be he nex e olu ion s ep in mobili y analysis using call de ail eco ds. All me hods and c ea ed and implemen ed algo i hms ha e he o esigh o easy adap a ion o u u e echnologies allowing con inua ion wo k o be ca ied ou . Au ho Con ibu ions: Concep ualiza ion, Ana Al es, Ma co Veloso and Ca los Ben o; o mal analysis, Gonçalo Fe ei a; in es iga ion, Gonçalo Fe ei a; me hodology, Gonçalo Fe ei a and Ana Al es; p ojec adminis a ion, Ca los Ben o; esou ces, Ca los Ben o; so wa e, Gonçalo Fe ei a; supe ision, Ana Al es, Ma co Veloso and Ca los Ben o; alida ion, Gonçalo Fe ei a; isualiza ion, Gonçalo Fe ei a; w i ing—o iginal d a , Gonçalo Fe ei a; w i ing— e iew and edi ing, Ana Al es and Ma co Veloso. All au ho s ha e ead and ag eed o he published e sion o he manusc ip . Funding: This esea ch ecei ed no ex e nal unding. Ins i u ional Re iew Boa d S a emen : No applicable. In o med Consen S a emen : No applicable. Da a A ailabili y S a emen : Res ic ions apply o he a ailabili y o he da a. Da a we e ob ained om a hi d pa y and a e a ailable om he au ho s wi h he pe mission o said hi d pa y. Con lic s o In e es : The au ho s decla e no con lic o in e es . Abb e ia ions The ollowing abb e ia ions a e used in his manusc ip : API Applica ion P og amming In e ace CDRs Call De ail Reco ds DBSCAN Densi y-Based Spa ial Clus e ing o Applica ions wi h Noise EVPs Excep ionally Visi ed Places GPS Global Posi ioning Sys em MVPs Mos Visi ed Places OVPs Occasionally Visi ed Places POIs Poin s o In e es ISPRS In . J. Geo-In . 2022,11, 228 17 o 18 Re e ences 1. Gonzalez, M.C.; Hidalgo, C.; Ba abasi, A.L. Unde s anding Indi idual Human Mobili y Pa e ns. Na u e 2008 ,453, 779–782. [C ossRe ] [PubMed] 2. Gu, Q.; Sacha idis, D.; Ma hioudakis, M.; Wang, G. In e ing Venue Visi s om GPS T ajec o ies. In P oceedings o he 25 h ACM SIGSPATIAL In e na ional Con e ence on Ad ances in Geog aphic In o ma ion Sys ems, Redondo Beach, CA, USA, 7–10 No embe 2017; Associa ion o Compu ing Machine y: New Yo k, NY, USA, 2017. [C ossRe ] 3. P oux, D.; Roulland, F. Mobile Recommenda ion Challenges wi hin a S ong P i acy O ien ed Pa adigm. In P oceedings o he 3 d ACM SIGSPATIAL In e na ional Wo kshop on Loca ion-Based Recommenda ions, Geosocial Ne wo ks and Geoad e ising, Chicago, IL, USA, 2019; Associa ion o Compu ing Machine y: New Yo k, NY, USA, 5–8 No embe 2019. [C ossRe ] 4. Zhang, D.; Huang, J.; Li, Y.; Zhang, F.; Xu, C.; He, T. Explo ing Human Mobili y wi h Mul i-Sou ce Da a a Ex emely La ge Me opoli an Scales. In P oceedings o he 20 h Annual In e na ional Con e ence on Mobile Compu ing and Ne wo king, Maui, HI, USA, 7–11 Sep embe 2014; Associa ion o Compu ing Machine y: New Yo k, NY, USA, 2014; pp. 201–212. [C ossRe ] 5. Ranjan, G.; Zang, H.; Zhang, Z.L.; Bolo , J. A e Call De ail Reco ds Biased o Sampling Human Mobili y? Mob. Compu . Commun. Re . 2012,16, 33–44. [C ossRe ] 6. Vanhoo , M.; Reis, F.; Smo eda, Z.; Ploe z, T. De ec ing home loca ions om CDR da a: In oducing spa ial unce ain y o he s a e-o - he-a . a Xi 2018, a Xi :1808.06398. 7. Ayesha, B.; Jeewan hi, B.; Chi a anjan, C.; Pe e a, A.S.; Kuma age, A.S. Use Localiza ion Based on Call De ail Reco d. In In elligen Da a Enginee ing and Au oma ed Lea ning–IDEAL 2019; Yin, H., Camacho, D., Tino, P., Tallón-Balles e os, A.J., Menezes, R., Allmendinge , R., Eds.; Sp inge In e na ional Publishing: Cham, Swi ze land, 2019; pp. 411–423. 8. Yang, P.; Zhu, T.; Wan, X.; Wang, X. Iden i ying Signi ican Places Using Mul i-Day Call De ail Reco ds. In P oceedings o he 2014 IEEE 26 h In e na ional Con e ence on Tools wi h A i icial In elligence, Washing on, DC, USA, 10–12 No embe 2014; pp. 360–366. [C ossRe ] 9. Isaacman, S.; Becke , R.; Cáce es, R.; Kobou o , S.; Ma onosi, M.; Rowland, J.; Va sha sky, A. Iden i ying Impo an Places in People’s Li es om Cellula Ne wo k Da a. In Pe asi e Compu ing; Lyons, K., High owe , J., Huang, E.M., Eds.; Sp inge : Be lin/Heidelbe g, Ge many, 2011; pp. 133–151. 10. Quad i, C.; Zignani, M.; Gai o, S.; Rossi, G.P. On Non-Rou ine Places in U ban Human Mobili y. In P oceedings o he 2018 IEEE 5 h In e na ional Con e ence on Da a Science and Ad anced Analy ics (DSAA), Tu in, I aly, 1–3 Oc obe 2018; pp. 584–593. [C ossRe ] 11. Qian, C.; Li, W.; Duan, Z.; Yang, D.; Ran, B. Using mobile phone da a o de e mine spa ial co ela ions be ween ou ism acili ies. J. T ansp. Geog . 2021,92, 103018. [C ossRe ] 12. Csáji, B.C.; B owe , A.; T aag, V.; Del enne, J.C.; Huens, E.; Van Doo en, P.; Smo eda, Z.; Blondel, V.D. Explo ing he mobili y o mobile phone use s. Phys. A S a . Mech. Appl. 2013,392, 1459–1473. [C ossRe ] 13. Hess, A.; Ma sh, I.; Gillblad, D. Explo ing communica ion and mobili y beha io o 3G ne wo k use s and i s empo al consis ency. In P oceedings o he 2015 IEEE In e na ional Con e ence on Communica ions (ICC), London, UK, 8–12 June 2015; pp. 5916–5921. [C ossRe ] 14. Leno mand, M.; Pico nell, M.; Can ú-Ros, O.G.; Tugo es, A.; Louail, T.; He anz, R.; Ba helemy, M.; F ías-Ma ínez, E.; Ramasco, J.J. C oss-Checking Di e en Sou ces o Mobili y In o ma ion. PLoS ONE 2014,9, e105184. [C ossRe ] [PubMed] 15. Ding, J.; Ni, C.C.; Gao, J. Figh ing S a is ical Re-Iden i ica ion in Human T ajec o y Publica ion. In P oceedings o he 25 h ACM SIGSPATIAL In e na ional Con e ence on Ad ances in Geog aphic In o ma ion Sys ems, Redondo Beach, CA, USA, 7–10 No embe 2017; Associa ion o Compu ing Machine y: New Yo k, NY, USA, 2017. [C ossRe ] 16. Vancea, F.; Vancea, C.; Popescu, D.; Zma anda, D.; Gabo , G. Secu e Da a Re en ion o Call De ail Reco ds. In . J. Compu . 2010 , 5, 961–967. [C ossRe ] 17. Yang, J.; Dash, M.; Teo, S. PPTPF: P i acy-P ese ing T ajec o y Publica ion F amewo k o CDR Mobile T ajec o ies. ISPRS In . J. Geo In . 2021,10, 224. [C ossRe ] 18. Khos ow-Pou , D.B.A. Encyclopedia o In o ma ion Science and Technology, 3 d ed.; IGI Global: He shey, PA, USA, 2015. [C ossRe ] 19. Phi hakki nukoon, S.; Ho anon , T.; Di Lo enzo, G.; Shibasaki, R.; Ra i, C. Ac i i y-Awa e Map: Iden i ying Human Daily Ac i i y Pa e n Using Mobile Phone Da a. In Human Beha io Unde s anding; Salah, A.A., Ge e s, T., Sebe, N., Vincia elli, A., Eds.; Sp inge : Be lin/Heidelbe g, Ge many, 2010; pp. 14–25. 20. Zhou, X.; Liu, J.; Yeh, A.G.O.; Yue, Y.; Li, W. The Unce ain Geog aphic Con ex P oblem in Iden i ying Ac i i y Cen e s Using Mobile Phone Posi ioning Da a and Poin o In e es Da a. In Ad ances in Spa ial Da a Handling and Analysis: Selec Pape s om he 16 h IGU Spa ial Da a Handling Symposium; Sp inge In e na ional Publishing: Cham, Swi ze land, 2015; pp. 107–119. [C ossRe ] 21. Wang, F.; Chen, C. On da a p ocessing equi ed o de i e mobili y pa e ns om passi ely-gene a ed mobile phone da a. T ansp. Res. Pa C Eme g. Technol. 2018,87, 58–74. [C ossRe ] [PubMed] 22. Diao, M.; Zhu, Y.; Fe ei a, J.; Ra i, C. In e ing indi idual daily ac i i ies om mobile phone aces: A Bos on example. En i on. Plan. Plan. Des. 2015,43, 920–940. [C ossRe ] 23. Yuan, G.; Chen, Y.; Sun, L.; Lai, J.; Li, T.; Zhuo, L. Recogni ion o Func ional A eas Based on Call De ail Reco ds and Poin o In e es Da a. J. Ad . T ansp. 2020,2020, 1–16. [C ossRe ] 24. A ailable online: h ps://www.po ugal.go .p /p /gc22/go e no/comunicados-do-conselho-de-minis os?p=13 (accessed on 13 Feb ua y 2021). ISPRS In . J. Geo-In . 2022,11, 228 18 o 18 25. He, R.; Cao, J.; Zhang, L.; Lee, D. S a is ical En ichmen Models o Ac i i y In e ence om Imp ecise Loca ion Da a. In P oceedings o he IEEE INFOCOM 2019-IEEE Con e ence on Compu e Communica ions, Pa is, F ance, 29 Ap il–2 May 2019; pp. 946–954. [C ossRe ] 26. And ade, R.; Al es, A.; Ben o, C. POI Mining o Land Use Classi ica ion: A Case S udy. In . J. Geo In . 2020,9, 493. [C ossRe ] 27. Io an, C.; Ol eanu-Raimond, A.M.; Cou onné, T.; Smo eda, Z. Mo ing and Calling: Mobile Phone Da a Quali y Measu emen s and Spa io empo al Unce ain y in Human Mobili y S udies. In Geog aphic In o ma ion Science a he Hea o Eu ope; Sp inge In e na ional Publishing: Cham, Swi ze land, 2013; pp. 247–265. [C ossRe ]