A Short-term model for detecting high traffic speed violation by using machine learning approach

Speed violation is one of the main causes of accidents. High speed not only increases the probability of occurrence of accidents but also increases the severity of accidents. So a vital point is trying to prevent the occurrence of speed violations. In this study, the hourly average traffic speed for Khorramabad to Arak highway is predicted for the future. If the predicted speed is near or exceeds the permitted speed, it is necessary to consider arrangements and preparations to reduce the average speed of traffic by users or the transportation network operators. In order to predict hourly average traffic speed, related traffic data was recorded in recent years. Many new features that affect traffic speed are extracted and used in predictive models. Three machine learning methods, including support vector machine, artificial neural network, and recurrent neural network, have been used. All three methods have the ability to analyze big traffic data, and in addition, the recurrent neural network has more consistency with the time-series nature of data. The results show that for both directions of this highway, by using only calendar and weather features, the mean absolute percentage error of the models is varied between 2.8 to 5.1 percent. Models can predict speeds over 80 kilometers per hour with precision over 80 percent. By adding the observed speed of the previous 3 to 8 hours as predictive features, the mean absolute percentage error of the models is decreased to 2.5 to 4.6.