In recent years, increasing urban mobility and complex traffic dynamics have intensified the need for accurate traffic flow forecasting in intelligent transportation systems. However, existing models often struggle to jointly capture short-term fluctuations and long-term temporal dependencies under noisy and heterogeneous traffic conditions. To address this challenge, this paper proposes a hybrid traffic flow forecasting framework that integrates Density-Based Spatial Clustering of Applications with Noise (DBSCAN), the Mamba state-space model, and the Transformer architecture. The framework first applies DBSCAN to multidimensional traffic features to enhance traffic state representation and reduce noise. The prediction module alternates between MambaBlocks, which capture local dynamics and high-frequency variations via state-space modeling, and TransformerBlocks, which employ a dynamic gating mechanism to efficiently model long-range dependencies using multi-query and linear attention. Experimental results demonstrate that the proposed model consistently outperforms baseline methods, including Decomposition-Linear (DLinear), Autoformer, and the standard Transformer, across 5-, 10-, 30-, and 60-min prediction horizons in terms of mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Ablation studies further verify the complementary contributions of DBSCAN preprocessing, Mamba-based short-term modeling, and Transformer-based global context learning. Overall, the proposed framework provides an effective and scalable solution for traffic flow forecasting in complex and dynamic environments.
Zhang et al. (Thu,) studied this question.