Modeling and Simulation of Human Choices: from Utility Theory to - PowerPoint PPT Presentation

Modeling and Simulation of Human Choices: from Utility Theory to Applications Prof. Michel Bierlaire Director – Transportation Center Ecole Polytechnique Fédérale de Lausanne (EPFL) Switzerland

Introduction : Science Fiction  Psyc Psychohistory hohistory: B : Bra ranc nch of h of mathe thematic tics whic s which de h deals ls with the with the re reactions of hum tions of human conglom onglomera rate tes to fixe s to fixed d soc socia ial a l and e nd econom onomic ic stim stimuli. Enc uli. Encyc yclope lopedia dia Ga Gala lactic tica, 1 , 116th Edition th Edition (1 (1020 F.E.) F.E.)

Introduction: Prof. McFadden  La Laure ureate te of The of The B Bank nk of of Swe Swede den Prize n Prize in Ec in Econom onomic ic Sc Scie ienc nces in Me s in Memory of ory of Alfre lfred N d Nobe obel 2 l 2000  Owns a Owns a fa farm rm a and vine nd vineya yard in rd in Napa pa Va Valle lley y  “ Fa Farm rm work work c cle lears the rs the m mind, ind, and the nd the vine vineya yard is a rd is a gre great t pla place to prove to prove the theore orems ” ”

Introduction : marketing  Pre Predic diction of tion of mark rket sha t share res s  Choic hoice of bra of brand nd  Choic hoice of produc of product t fe feature tures s  Choic hoice of re of reta tail il store store  Etc Etc. .

Introduction : transportation demand analysis  Choic hoice of m of mode ode  Choic hoice of pa of path th  Choic hoice of of de destina stination tion  Choic hoice of pa of park rking ing  Choic hoice of of de depa parture rture tim time

Framework Data Model Simulation

Data: questionaires  Data ta a about the bout the re responde spondent nt  Choic hoice da data ta  Reve veale led d pre prefe fere renc nces s  Sta State ted pre d prefe fere renc nces

Data: smartphones  GSM, GPS GSM, GPS  Accele lerom romete ter r  WiFi WiFi  Blue luetooth tooth  Ambie bient sound nt sound  And m nd more ore... ...

Data: scanner data  Deta taile iled purc d purcha hase se inform information tion  Pe Persona rsonalize lized d

Data: eye tracking  Whe Where re do pe do people ople look look?  Use sed in m d in mark rketing ting re rese searc rch h  Use sed in driving d in driving sa safe fety re ty rese searc rch h  R Rele leva vant for nt for pe pede destria strian m n mode odels ls

Model : assumptions  Hom omo e o econom onomic icus us  Rationa tionality lity  Utility the tility theory ory  Ea Each a h alte lterna rnative tive is is assoc ssocia iate ted with a d with a utility utility  The The a alte lterna rnative tive with the with the la large rgest utility is c st utility is chose hosen n

Model : assumptions  Strong Strong assum ssumptions ptions  Unc ncerta rtainty a inty and nd irra irrationa tionality m lity must ust be c be capture ptured d  Random ndom utility utility mode odels ls  La Late tent va nt varia riable bles s

Model : features  Disa isaggre ggrega gate te – – mark rket se t segm gments nts  Qua Quantita ntitative tive a and nd qua qualita litative tive va varia riable bles s  Can ha n handle ndle subje subjectivity - tivity - attitude ttitudes- s- pe perc rceptions ptions

Application : simulation of market shares  Polic Policy va y varia riable bles s (e (e.g. pric .g. price) )  Nonline onlinear e r effe ffect t

Application : market segmentation  Ma Mark rket sha t share res pe s per r se segm gment nt  Gra Granula nularity rity de depe pends on the nds on the da data ta a ava vaila ilability bility

Application : simulation of revenues  Conc oncept of optim pt of optimal l pric price  Can be n be se segm gment nt spe specific ific

Application : pedestrian walking behavior  Choic hoice of the of the ne next xt ste step p  Collision a ollision avoida voidanc nce  Le Leade der followe r follower r

Application : pedestrian simulation

Application : pedestrian simulation

Applications: route choice  Com omple plex proble x problem  Num umbe ber of pa r of paths is ths is huge huge  High le igh leve vel of l of ove overla rlapping pping  Shorte Shortest pa st path not th not be beha haviora viorally lly meaningful ningful

Application : electric vehicles  Ma Mark rket sha t share res s  Hypothe ypothetic tical l choic hoice  Im Importa portanc nce of of attitude ttitude towa toward the rd the environm nvironment nt

Application : facial expression recognition  Autom utomatic tic ide identific ntification of the tion of the emotion otion  Pote Potentia ntially lly diffe differe rent a nt across ross culture ultures s  Require quires a s adva dvanc nced im image ge proc processing ssing algorithm lgorithms s

Application : demand-supply interactions  Reve venue nue mana nage gement nt  Ma Mark rket e t equilibrium quilibrium  Com ombina bination of tion of ope opera rations tions re rese searc rch a h and nd de demand m nd mode odels ls

Conclusion  Disc iscre rete te c choic hoice m mode odels ls  Adva dvanc nced a d and ope nd opera rationa tional l  Accom omoda odate te m mode odern da rn data ta sourc sources s  Wide Wide ra range nge of a of applic pplications tions  Com omple plex m x mode odels re ls require quires sim s simula ulation tools tion tools

Short course : Discrete Choice Analysis: Predicting Demand and Market Shares  Janua nuary 2 ry 29- Fe - Februa bruary 2 ry 2, , 2012 2012  Ec Ecole ole Polyte Polytechnique hnique Fé Fédé déra rale le de de La Lausa usanne nne  Prof. B Prof. Ben-A n-Akiva iva (MIT) – (MIT) – Prof. Bie Prof. B ierla rlaire ire (EPFL) (EPFL)  tra transp-or.e nsp-or.epfl.c pfl.ch/ h/dc dca