Freeway incident detection using hybrid fuzzy neural network

Abstract

A new technique for freeway incident detection using a hybrid neuro-fuzzy system is proposed. This neuro-fuzzy system uses a self-rule-generating algorithm that organises the training data into clusters and learns the fuzzy rules automatically. Gaussian membership functions are assigned based on statistical properties of the training data set. Fuzzy rules are automatically obtained from the clusters and a neural network constructed using them. The training of parameters is performed using two modified linear least squares regression models. Different algorithms are implemented to obtain improvement in the speed of convergence. Real I-880 freeway traffic data are used to test the effectiveness of the developed fuzzy-neural system. To assess the transferability of the trained system, the network was trained on AYE dataset from Singapore and then adapted onto I-880 dataset from USA. The system is observed to be highly adaptable giving excellent results after adaptation. The results obtained show high potential for the application of this neuro-fuzzy system to the problem of freeway traffic incident detection.