AST-MTL: An Attention-Based Multi-Task Learning Strategy for Traffic Forecasting

Road traffic forecasting is crucial in Intelligent Transportation Systems (ITS). To achieve accurate results, it necessary to model the dynamic nature and complex non-linear dependencies governing traffic. The goal particularly challenging when prediction involves more than just one variable. This paper proposes a novel multi-task learning model, called AST-MTL, perform multi-horizon predictions of flow speed at road network scale. strategy combines multilayer fully-connected neural (FNN) multi-head attention mechanism learn related tasks while improving generalization performance. also includes graph convolutional (GCNs) gated recurrent unit (GRUs) extract spatial temporal features conditions. Our experiments employ new sets GPS data, OBU freeway urban contexts. experimental results prove our can effectively for different types roads outperform state-of-the-art models.