Rough Sets and FCM-Based Neuro-fuzzy Inference System for Traffic Incident Detection

Detecting incidents on urban freeway or arterials using loop detector data is quite challenging. Considering the ability of fuzzy clustering for data discretization, that of rough sets theory to reduction of decision system, and that of fuzzy neural networks to nonlinear mapping, a novel hybrid neuron-fuzzy inference method that synergies fuzzy c-means (FCM),rough sets theory, and adaptive neuro-fuzzy inference system for incident detection was proposed. Firstly, the continuous attributes from detector loop were discretized with FCM clustering. Then, attribute reduction were performed based on rough sets theory using generic algorithm, and the key conditions for incident were determined. Lastly, according to the chosen attribute data, the ANFIS(Adaptive-Network-Based Fuzzy Inference System) was designed for detection. The major advantage of this approach is to optimize the overall structure of ANFIS and avoid the "dimensional disaster" with rough sets theory to attribute reduction. The efficiency of the new method is also illustrated by means of applying to real traffic data, the result of the experiment demonstrated that the solution was very effective to increase the recognition rate and to reduce the number of false detections.