Flex Scheduling for Bus Arrival Time Prediction

Abstract

The prediction of bus arrival times is an important element for travel planning. This study used three weeks of Chicago, Illinois, Transit Authority bus route GPS data to compare the performance of several commonly used methods and algorithms. The use of implicit schedules in previous papers was inadequate. The use of additional information, such as recent travel times along the route, is unnecessary. In addition, the use of computationally intensive machine learning algorithms, such as support vector regression, k nearest neighbor regression, and neural networks, is unnecessary. The study used basis expansion functions at various resolutions with linear models and cross-validated the models to determine explicitly the approximate historical interstop travel times for any time of the day and any day of the week. Combining the estimated interstop travel times with the real-time GPS location of a bus resulted in a flex schedule that was independent of scheduled departure or arrival times. Using a flex schedule makes the use of additional GPS information or the use of the machine learning algorithms unnecessary.