Long-term Traffic Speed Estimation via Regression using Weekly Day Patterns

Traffic is one of the biggest problem of the city-life. Inadequate roads, the increasing number of people and vehicles, inefficient transportation plans and many other reasons cause people to spend long time at traffic. There are several mobile and web applications which offer alternative shortest routes based on instant traffic information in order to minimize the spent time during the journeys. The biggest disadvantage of these applications is that they are not able to make a reliable prediction of the traffic flow rate of a given hour or date. In this study, we propose a simple regression model which is based on the patterns of the past k weeks to predict the traffic flow of the following 7 days. The results of the proposed model outperform the predictions that are performed by exploiting mean values of the past weeks by 3%. Analyzing the traffic flow data of Istanbul which is collected in 2017, it is revealed that the proposed regression model could decrease the error rate of 1 to 7 days predictions down to 5%.