A Hybrid Model for Short-Term Flow Prediction Incorporating Spatial-Temporal Correlation in Road Networks

Predicting traffic flows is crucial for traffic management. Due to the complex and interconnected nature of road networks, traffic in an area is influenced not only by the temporal historical traffic, but also by the spatial traffic patterns in neighboring areas. This paper presents a hybrid methodology that incorporates spectral and spatial-temporal correlation analysis for each road segment and uses a boosting model for prediction (SST-Boost). Spectral analysis is utilized to decompose the traffic flow data into periodic trends and volatility components, and road segment relevance matrix is introduced to determine the input features for each road segment’s prediction model. By evaluating the data set collected from Yuhang District, Hangzhou, China, the proposed SST-Boost outperforms several commonly used models in traffic flow prediction, verifying the effectiveness of adopting the spectral analysis and spatial-temporal modules.