Meteorological Variables and Prediction of Road Traffic Accident Severity in Suzhou city of Anhui Province of China

Background: The prediction of the severity of traffic accidents is concerned by researchers and law enforcement. In order to simulate the relationship between road severity results and meteorological factors, a large number of models have been proposed. This study purpose is to conduct a machine learning model to investigate the impact of meteorological variables on the severity of road traffic accidents. Methods: Using data from the 2007 and 2008 -2017 the Traffic Police Detachment of the Public Security Bureau of Suzhou, 7,795 traffic accidentswere included in this study. We attempted to use a random forest model to convey the nonlinear relationship between meteorological variables and the severity of traffic accidents, and to compare the prediction accuracy of the neural network model. The model is constructed by the randomForest package and the neuralnet package in the R software. 75% of the training samples were divided from the data to establish a prediction model, and the remaining 25% of the test samples were used for testing. In addition, in order to understand the accuracy of the model prediction, the predicted results were calculated and compared with the actual results. Results: In the random forest model, the most optimal mtry parameter value was 5, the number of decision trees is 400. The weight of wind direction, atmospheric pressure and temperature might be higher than other variables. The OOB (out of bag) estimate of error rate was 51.09%, and the error rate for general traffic accident prediction is the lowest (45.97%). Similarly, in the neural network model, the calculated error rate is 61.01%, with the lowest error rate for minor traffic accidents (35.84%). Conclusions: The results of this study show that how using meteorological data predicts the severity of a traffic accident with relative accuracy, and the random forest model may be more suitable than the neural network model. Research and application of machine learning algorithms in the field of traffic accidents should be further explored.