Application of Neural Network in Modeling Commuter Choice Behavior with a Novel Fuzzy Access Measure

Public transit systems (PTS) in major cities of developing countries are under immense pressure due to the rising populations. Most city dwellers in these countries depend on PTS for their daily commute and are highly sensitive to shifting to private vehicles when utility of the PTS decreases for them. Accessibility to PTS is an aspect of transport planning which impacts the shifting behavior of the commuters to a large extent. For assessing the impact of access to PTS on commuter choice behavior; binary logistic regression and neural network (NN) models were developed using socio demographic and commute related data collected through a revealed preference (RP) survey experiment in Indore city in India. Accessibility to the PTS was quantified using a fuzzy weighted average (FWA) measure employing temporal impedances experienced by the commuters and the value they associate with each of the impedance. Both models show significant correlation between quantified accessibility and choice behavior. The neural network model developed for this study shows better predictive powers as the apparent validity of the model was found to be 0.883 (area under curve for the receiver’s operational characteristic curve) which was greater as compared to apparent validity for regression model, 0.753. Predictions made by the models for shifting behavior were validated using data from a stated preference (SP) survey experiment for the same sample set. The NN model performed better in this scenario as well with 68.34% correct predictions as opposed to 65.96% correct predictions from the regression model.