Short-Term Forecasting of Transit Travel Demand: A Customized Relevance Vector Machine Model

Abstract

Accurate forecasting of short-term travel demand is essential for the development of intelligent transportation systems. This paper studies the short-term forecasting of transit travel demand by proposing a customized relevance vector machine (C-RVM) model. The proposed C-RVM model takes advantage of the conventional RVM model but incorporates two data preprocessing sectors that adapt to the historical process changes and capture the dynamic information of the data. The historical travel demand data from two transit systems, an urban rail transit system and a bus transit system, are employed to evaluate the forecasting performance of the proposed C-RVM model. The results show that the proposed C-RVM model outperforms several benchmark forecasting models with higher accuracy. Specifically, the root mean square error for the proposed C-RVM model is decreased by 61.68%, 55.54%, 40.97%, and 14.00%, respectively, in comparison with that for the Gaussian process regression, support vector machine, artificial neural network, and conventional RVM. Instead of only forecasting travel demand with deterministic outputs, the proposed C-RVM model provides a probability for each possible forecasting output, which provides informative insights into the management and operation of transit systems considering uncertainty.