Hybrid Neuro-Fuzzy Application in Short-Term Freeway Traffic Volume Forecasting

Abstract

A hybrid neuro-fuzzy application for short-term freeway traffic volume forecasting was developed. The hybrid model consists of two components: a fuzzy C-means (FCM) method, which classifies traffic flow patterns into a couple of clusters, and a radial-basis-function (RBF) neural network, which develops forecasting models associated with each cluster. The new hybrid model was compared with previously developed clustering-based RBF models. In addition, the dynamic linear model was studied for comparison. The study results showed that the clustering-based hybrid method did not produce time-lag phenomena, whereas the dynamic linear model and the RBF model without clustering revealed apparent time-lag phenomena. The forecasting performance for freeway traffic volumes from the San Antonio, Texas, TransGuide system shows that even though the hybrid of the FCM and RBF models appears to be promising, additional research efforts should be devoted to achieving more reliable traffic forecasting.