Traffic Flow Prediction Using SVR-Ant Colony Optimization: A Practical Case of Tehran Highway

Traffic estimation is one of the most important issues in traffic control parlances. In this work, an integrated approach is proposed, which is a combination of three algorithms including K-means clustering, Support Vector Regression (SVR) and Ant Colony Optimization (ACO) approach. Using the K-means clustering algorithm allows obtaining optimal values for SVR via ACO algorithm and then employ it to predict traffic flow. To carry out simulations, two realistic cases of traffic flow prediction for Tehran-Karaj and Tehran-Damavand highways is investigated at two checkpoints in the morning and afternoon periods. Further, to evaluate the quality of solutions obtained from the proposed method, a time series model was used to end comparisons. Based on the results, the NRMSE forecast error for the proposed method presents less as opposed to well-known SARIMA method for morning and evening periods. Therefore, the proposed method outperforms SARIMA in terms of prediction error; that is by 0.26 versus 0.31 and 0.11 versus 0.18 for Tehran-Karaj highway during the morning and evening intervals. According to the results for two main highways, the proposed method exhibits its suitability for practical application in traffic prediction with accurate solution and simplicity of application in real cases.