Improving traffic flow forecasting with relevance vector machine and a randomized controlled statistical testing

High-accuracy traffic flow forecasting is vital to the development of intelligent city transportation systems. Recently, traffic flow forecasting models based on the kernel method have been widely applied due to their great generalization capability. The aim of this article is twofold: A novel kernel learning method, relevance vector machine, is employed to short-term traffic flow forecasting so as to capture the inner correlation between sequential traffic flow data, it is a type of nonlinear model which is accurate and using only a small number of relevant basis functions automatically selected. So that it can find concise data representations which are adequate for the learning task retaining as much information as possible. On the other hand, the sample size for learning has a significant impact on forecasting accuracy. How to balancing the relationship between the sample size and the forecasting accuracy is an important research topic. A randomized controlled statistical testing is layout to evaluating the impacts of sample size of the new proposed traffic flow forecasting model. The experimental results show that the new model achieves similar or better forecasting and generalization performance compared to some old ones; besides, it is less sensitive to the size of learning sample.