Performance of Automatic ANN-Based Incident Detection on Freeways

Abstract

Automatic incident detection on freeways is an essential ingredient for the successful deployment of Intelligent Transportation Systems. Several incident detection algorithms have been developed in the past three decades; however, most of them have not shown the anticipated performance in terms of detection rate and false alarm rate. Recently, the artificial neural networks (ANN) have been introduced to incident detection and shown success over the traditional algorithms. This study explores the application of two neural network models, namely, the Multi-Layer Feed-Forward and the Fuzzy ART algorithm. This study was conducted on the central corridor of I-4 in Orlando using real-world data collected via the traffic surveillance system. Different scenarios were considered to improve the performance and to capture the sensitivity of the developed algorithms to some factors. The study results showed that the Fuzzy ART algorithm has generally outperformed the Multi-Layer Feed-Forward network and California algorithms #7 and #8.