Assessing the performance of artificial neural network incident detection models

This paper explores the performance of a relatively new-generation of algorithms for automated freeway incident detection using Artificial Neural Networks (ANNs). These new models have the potential to provide faster and more reliable incident detection times and fault-tolerant operation while being easy to implement on existing and new hardware platforms. The ANN incident detection models were trained on data obtained from two freeways in Melbourne, Australia. Two sources of data were used to assemble the training data sets. The first comprised speed, flow and occupancy data from dual-loop detector stations and the second was an incident log showing the approximate time of incidents. This resulted in a database comprising a set of 100 incidents of varying severity and under a variety of traffic conditions at different locations on the freeway. Sixty of these incidents were used for training the ANN model while the remaining forty incidents were used for model testing. The model developed in this study uses speed, flow and occupancy data measured at dual stations, averaged across all lanes and only from one preceding time interval. The off-line performance of the model is reported under both incident and non-incident conditions. The incident detection performance of the model is reported based on a validation-test data set of forty incidents that were independent of the sixty incidents used for training. The false alarm rates of the model are evaluated based on non-incident data that were collected from a freeway section which was video-taped for a period of 33 days. The results presented in this paper provide a comprehensive evaluation of the performance of the ANN model and confirm that neural network models can provide fast and reliable incident detection on freeways