Spatiotemporal Forecasting of Traffic Flow Using Wavelet-Based Temporal Attention

• Published in: IEEE access 2024, Vol.12, p.188797-188812
• Main Authors: Jakhmola, Yash, Panja, Madhurima, Kumar Mishra, Nitish, Ghosh, Kripabandhu, Kumar, Uttam, Chakraborty, Tanujit
• Format: Article
• Language: English
• Subjects: computational complexity; Correlation; Data models; Deep learning; Forecasting; Predictive models; Proposals; Roads; spatiotemporal data; Spatiotemporal phenomena; temporal attention; Time series analysis; Traffic forecasting; Urban areas; wavelet transformation
• ISSN: 2169-3536
• DOI: 10.1109/ACCESS.2024.3516195

Spatiotemporal forecasting of traffic flow data represents a typical problem for urban traffic management, involving complex interactions, nonlinearities, and long-range dependencies due to the interwoven nature of the temporal and spatial dimensions. Traditional statistical and machine learning methods struggle to handle both temporal and spatial dependencies in such datasets. While graph convolutional networks and multi-head attention mechanisms have been widely adopted in this field, they often fail to accurately model dynamic temporal patterns and effectively differentiate noise from signals in traffic datasets, leading to potential overfitting. This paper proposes a wavelet-based temporal attention model, namely a wavelet dynamic spatiotemporal aware graph neural network (W-DSTAGNN), for tackling the traffic forecasting problem. A key feature of W-DSTAGNN is the use of wavelets that can effectively separate noise and capture multi-resolution temporal patterns. Wavelet decomposition can help by decomposing the signal into components that can be analyzed independently, reducing the impact of non-stationarity and handling long-range dependencies of traffic flow datasets. These enable our proposal to generate more robust and accurate forecasts of complex and dynamic traffic dependencies than commonly used spatiotemporal deep learning models. Benchmark experiments using three popularly used statistical metrics confirm that our proposal efficiently captures spatiotemporal correlations and outperforms eleven state-of-the-art models (including both temporal and spatiotemporal benchmarks) on three publicly available traffic datasets. Our proposed approach can better handle dynamic temporal and spatial dependencies, delivering reliable long-term forecasts, and generating interval forecasts to enhance probabilistic forecasting of traffic datasets.