Predicting the Traffic Crashes of Taxi Drivers by Applying the Non-Linear Learning of ANFIS-PSO with M5 Model Tree

As an essential issue in traffic crashes, human factor plays an indispensable role. This study uses the general health questionnaire (GHQ-12) within some socio-demographic and also a number of daily exercise related questions for prediction of traffic crashes among taxi drivers in the City of Tehran. A novel technique is been developed by applying nonlinear-learning of composition model of Adaptive Neuro-Fuzzy Inference Systems (ANFIS) and Particle Swarm Optimization (PSO) with M5 model tree. In order to improve the generalization ability of a single data driving algorithm, a cluster of ANFIS models with different nodes and hidden layers are implemented to extract the inherent relationship between traffic accident rates and human factors. The predictions of ANFIS models are combined applying a nonlinear weighted average method composed of M5 tree, and the PSO is utilized to find the optimal parameters of ANFIS models. The performance of the proposed model is evaluated in a case study and the relevant data are collected from a large group of Taxi drivers in the City of Tehran, Iran; as it was carried out to predict the associated accident rates. The Nash-Sutcliffe coefficient (NSE) and different error criteria are utilized to assess the prediction efficiency of the associated Hybrid model. Results indicate that the M5 tree is capable in enhancing the prediction accuracy of the novel model applied in the prediction of the accident rates as in comparison with the popular ANFIS model. Moreover, the linear relationships generated by M5 tree show the sensitivity of ensembled model accuracy on the single ANFIS models, which have a partial tendency in underestimation of the traffic crashes.