Traffic management via traffic parameters prediction by using machine learning algorithms

One of the short-term strategies to manage the traffic and make a balance between travel supply and demand for the near future is the short-term prediction of traffic parameters and informing the passengers. Therefore passengers are more likely to avoid traveling during traffic peak hours. In this study, hourly average traffic speed and hourly traffic volume as two traffic parameters that indicate traffic state are predicted for Karaj-Chaloos road in Iran. Since traffic data have large volume, machine learning-based models have more suitable performance than traditional models. However, it is not merely possible to discover the cause and effect relationships and the importance of features. In this study, after using the artificial neural network and K-nearest neighbor models to predict traffic parameters, to analyze the sensitivity of the results, the importance of used features is investigated. Also, the effect of passing the time over the accuracy of predictions has been examined. According to the results, the highest accuracy of predicting hourly traffic volume and hourly average traffic speed is achieved by the K-nearest neighbor that is equal to 61% and 91%, respectively. Compared to the historical average as a benchmark model, a significant improvement in the accuracy of predictions has been obtained by the artificial neural network and K-nearest neighbor models.