Forecasting Shared-Use Vehicle Trips with Neural Networks and Support Vector Machines

Abstract

Two trip-forecasting approaches—neural networks and support vector machines—are compared for a multiple-station shared-use vehicle system. The neural networks trained to perform trip forecasting belong to the multilayer perceptron model. Comparative evaluation was made with least-squares support vector machines with the radial basis kernel function. The forecasting models were trained or developed for 6 months, then validated with 1 month of actual trip data from the Honda Intelligent Community Vehicle System, currently in commercial operation in Singapore. Each model was designed to forecast the net flow of vehicles for a 3-h period on any day in a month at a particular shared-use vehicle port. The models were developed for and applied to forecasting the net flow at each port for each entire month from January 2004 to June 2004. Results indicate that the multilayer perceptron model has a slightly better forecast accuracy in terms of monthly average absolute error and monthly maximum absolute error.