Mobility is a fundamental element of modern society and the study of mobility demand plays a substantial role for decision makers in the development of transport supply and infrastructure. An important element in the analysis of mobility demand is the modal split, which represents the percentage of trips made using a given mode of transport. Estimating the modal split is nonetheless very complex and, based on the literature, the approaches used today have some significant limits: design complexity, low reliability of results, high costs and low scalability. The aim of this thesis is therefore to propose an alternative approach to solve some of these limits. To reach this objective, an analysis of the drivers influencing the choice of transport mode and of the models used to perform this analysis is firstly carried out. Then, from this study, a new model to predict the modal choice within the Lombardy Region is developed. The proposed model is based on a Multinomial Logistic Regression, which makes it possible to predict, for each route, the probability that an individual chooses one mean of transport over another, and to understand how each variable influences the final choice. To do this, the Origin-Destination Matrix of Lombardy Region (2014), integrated with data related to the municipalities of the region, is exploited. The obtained model shows a good level of performance, correctly allocating, on average, more than 86% of the trips, and performs better than two comparison models: a dummy classifier predicting only the prevailing class within the dataset and another model predicting only the average target values. In addition, the analysis of the coefficients of the model allows both the identification of the factors influencing the modal split, and the analysis of how a change in these modifies the probability of mode choice. In this sense, the relevant variables identified are: the population of the origin and destination municipality, the number of families per population in the municipality of destination, the number of cars per capita registered in the municipality of origin and destination, the number of public transport (buses and trolleybuses) per capita in the municipality of destination, the IRPEF per capita in the municipality of destination and the average travel time (travelling by car).
La mobilità è un elemento fondamentale della società moderna e lo studio della domanda di mobilità gioca un ruolo sostanziale per i decision makers nello sviluppo dell’offerta e dell’infrastruttura del trasporto. Un elemento rilevante dell’analisi della domanda di mobilità è costituito dal modal split, rappresentante la percentuale di viaggi effettuati utilizzando un determinato mezzo di trasporto. Effettuare una stima del model split è tuttavia molto complesso e, sulla base di quanto emerge dalla letteratura, gli approcci oggi utilizzati presentano alcuni limiti significativi: complessità di design, bassa attendibilità dei risultati, elevati costi e bassa scalabilità. La tesi si pone quindi l’obiettivo di proporre un approccio alternativo che consenta di risolvere alcuni di questi limiti. A tal fine, viene in primo luogo effettuata un’analisi dei driver che influenzano la scelta della modalità di trasporto e dei modelli utilizzati per effettuare tale analisi. A partire da tale analisi, viene sviluppato un nuovo modello volto a prevedere la scelta modale all’interno del territorio lombardo. Il modello proposto si basa su una Regressione Logistica Multinomiale che consente di predire la probabilità che un individuo scelga un mezzo di trasporto rispetto a un altro per ogni tratta, e comprendere come ciascuna variabile influenzi la scelta finale. Per fare ciò, viene utilizzata la matrice Origine-Destinazione di Regione Lombardia (2014) integrata con dati relativi ai comuni della regione. Il modello ottenuto mostra un buon livello di performance, allocando correttamente, in media, più dell'86% dei viaggi e performa meglio di due modelli di confronto: un dummy classifier che prevede solo la classe prevalente del dataset e un altro modello che prevede solo i valori target medi. Inoltre, l’analisi dei coefficienti del modello consente sia di individuare i fattori influenzanti il modal split, sia di analizzare come un cambiamento in questi modifichi le probabilità di scelta della modalità di trasporto. In tal senso, le variabili rilevanti individuate sono: la popolazione del comune di origine e destinazione, il numero di nuclei famigliare pro capite del comune di destinazione, il numero di autovetture pro capite registrate nell’origine e destinazione, il numero di mezzi pubblici (autobus e filobus) pro capite nel comune di destinazione, l’IRPEF pro capite della destinazione e il tempo di percorrenza medio (spostandosi in auto).
Development of a new approach for understanding the modal split : the case of Lombardy Region
Decorato, Federica Beatrice;Dentici, Emanuele
2019/2020
Abstract
Mobility is a fundamental element of modern society and the study of mobility demand plays a substantial role for decision makers in the development of transport supply and infrastructure. An important element in the analysis of mobility demand is the modal split, which represents the percentage of trips made using a given mode of transport. Estimating the modal split is nonetheless very complex and, based on the literature, the approaches used today have some significant limits: design complexity, low reliability of results, high costs and low scalability. The aim of this thesis is therefore to propose an alternative approach to solve some of these limits. To reach this objective, an analysis of the drivers influencing the choice of transport mode and of the models used to perform this analysis is firstly carried out. Then, from this study, a new model to predict the modal choice within the Lombardy Region is developed. The proposed model is based on a Multinomial Logistic Regression, which makes it possible to predict, for each route, the probability that an individual chooses one mean of transport over another, and to understand how each variable influences the final choice. To do this, the Origin-Destination Matrix of Lombardy Region (2014), integrated with data related to the municipalities of the region, is exploited. The obtained model shows a good level of performance, correctly allocating, on average, more than 86% of the trips, and performs better than two comparison models: a dummy classifier predicting only the prevailing class within the dataset and another model predicting only the average target values. In addition, the analysis of the coefficients of the model allows both the identification of the factors influencing the modal split, and the analysis of how a change in these modifies the probability of mode choice. In this sense, the relevant variables identified are: the population of the origin and destination municipality, the number of families per population in the municipality of destination, the number of cars per capita registered in the municipality of origin and destination, the number of public transport (buses and trolleybuses) per capita in the municipality of destination, the IRPEF per capita in the municipality of destination and the average travel time (travelling by car).File | Dimensione | Formato | |
---|---|---|---|
Decorato_Dentici.pdf
accessibile in internet per tutti
Dimensione
4.85 MB
Formato
Adobe PDF
|
4.85 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/175838