Attention-Based Convolutional Recurrent Neural Network for Traffic Index Prediction

Abstract

Accurate and real-time prediction of traffic indexes are a basis for mastering the change of road network state, making traffic control measures, and travel planning reasonable. The ACRNN model which combines an attention mechanism, convolution neural network, and recurrent neural network is proposed to predict the traffic index with multiple factors. First, the eXtreme Gradient Boosting model is used to calculate the importance of features. Then, the ACRNN model is constructed, the spatial and temporal features of the traffic index are captured by a convolution neural network and long short-term memory, and an attention mechanism is introduced to capture the global change trend. The performances of ACRNN model are tested by using 2 months of data from Shenzhen. The results show that the proposed model performs well and the prediction accuracy reaches 89.6%, which is better than ARIMA models and other neural network models such as BP, LSTM, CNN, and CRNN.