Fuzzy logic application to model uncertain route choice behaviour of bus users

Abstract

Identifying the travelers’ route choice pattern is one of the important tasks of the transport planning. It is an important element for travel demand management of the existing transportation system and depends totally upon the individual’s decision. Such decision is uncertain in nature and depends on each user’s perception of different factors influencing route choice. The main objective of this research is to identify the bus users’ route choice pattern through two model approaches- Multinomial Logit Model and Fuzzy Logic Model. Multinomial Logit Model is one of the popular approaches for studying the route choice behavior. But as such decision of choosing any route involves uncertainty, Fuzzy Logic approach has been selected as the alternative method for addressing such uncertainty while modeling the route choice pattern. Two major bus stoppages of Dhaka city have been selected for conducting the study and user opinion survey namely Mirpur Sector 1 (origin point) and Motijheel Commercial Area (destination point). The factors which influence the route choice decision have been identified through users’ opinion survey and have been used as the independent variables in developing the models. The identified input variables are travel time, waiting time, travel cost, distance, comfort, safety, security, regularity, age, gender, and income level. The models have been developed for both weekday and weekend. It is to be noted that not all identified variables are statistically significant for developing the models. As such, numbers of models have been developed for different combination of input variables for both model approaches to identify the final models (for both weekday and weekend). In case of Multinomial Logit Model, the goodness-of-fit measures (Chi-Square Distribution, Log Likelihood results, Psedu R² value) have been compared for selecting the final model. As the notion of fuzzy logic is nonstatistical in nature and does not provide any goodness-of-fit measures, the output of each model (for different combination of input variables) have been compared with the actual field data for selecting the final models for both weekday and weekend. It has been found from the study that both model approaches (Multinomial Logit Model and Fuzzy Logic Model) have identified travel time and waiting time to be the significant factors for choosing the routes for both weekday and weekend. It has also been found that except Fuzzy Logic weekend model, the other models identified comfort, safety, security and regularity as important factors for choosing the routes. On the other hand, the comparison of the output results of both model approaches shows that Fuzzy Logic models can predict the route choice pattern (route share) more accurately than the Multinomial Logit Model for both weekday and weekend. This study proposes to include more origin and destination points to develop more precise and realistic model. The preciseness of the Fuzzy Logic approach depends on the quality of the field data. Therefore, it is recommended that more data needs to train the Fuzzy Logic models in Neuro-Fuzzy training which will result in more accurate results. This study concludes that Fuzzy Logic approach can be a better way for predicting the route share for its strength to address the uncertainty and impreciseness relationship between the input and output variables.