Short-Term Traffic Flow Prediction Based on the Efficient Hinging Hyperplanes Neural Network

Traffic flow (TF) prediction is an important and yet a challenging task in transportation systems, since the TF involves high nonlinearities affected by many elements. Recently, neural networks have attracted much attention for prediction, but they are commonly black boxes with complex architectures difficult to be interpreted, e.g., contributions of specific traffic elements not explicit, hardly providing informative guidance. In this paper, we aim at addressing more interpretable short-term joint consideration accuracy, thus introduces pragmatic method applying efficient hinging hyperplanes network (EHHNN) simply built upon sparse neuron connections. proposed method, different factors incorporated into inputs, including their spatial-temporal information. Besides pursuit further extend ANOVA decomposition EHHNNs interpretation analysis specifications data, which concerning variables detected quantitatively. As such, firstly applies EHHNN filter out dimensionality reduction while maintaining accurate prediction. Then, variable performed from perspectives, e.g. quantitatively investigate influence also impacts. Therefore, predictor analyzing tool can both attained exerting flexibility extending interpretability EHHNNs, promising provide guidance future control. Numerical experiments verify effectiveness potential analysis.