A Traffic Flow Prediction Method Using Road Network Data

Traffic flow prediction is one of the major tasks in urban intelligence. Various deep learning models have been proposed. However, the main interest was to capture spatial and temporal features; therefore, less data on external factors was used in the previous literature. This paper proposes a method of traffic flow prediction which can process road network data with two units: GIS subnet and mask processing layer. GIS subnet is a CNN-based subnet that can learn road network data, and the mask processing layer is placed at the bottom of the model to fit the prediction values. Experimental results with taxi flow data showed that the proposed method is effective in improving prediction accuracy.