Using Automatic Vehicle Location Data to Model and Identify Determinants of Bus Bunching

Bunching is recognized as one of the deteriorating factors affecting the performance of public transport networks. In this study, for the first time, Gene Expression Programming (GEP) and Decision Tree (DT) methods are utilized to estimate and model bus bunching. These methods are well equipped for dealing with nonlinearity and solving complex problems. The proposed models are compared against well-known Logistic Regression (LOR) models. Different spatial and temporal independent variables such as: bus dwell time, intersection delay, schedule deviation, bus stop spacing and bus stop closeness are used to model and study bus bunching in a real-life example in Auckland, New Zealand. Schedule deviation was determined to be the most influential factor for bus bunching occurrence. The DT method performed better in estimation of bus bunching occurrence compared to the GEP and LOR models. The LOR model inflates minor fluctuations and is prone to overestimation, reducing its predictive performance. (C) 2017 The Authors. Published by Elsevier B.V.