<p>This study presents the development of AST-Deep, an attention-based deep learning model aimed at improving the accuracy and reliability of short-term traffic flow forecasting. Traffic forecasting plays a critical role in managing road operations, reducing congestion, enhancing safety, and optimizing transportation resources. The AST-Deep model consists of three key steps: (1) Spatial Correlation Analysis, where an enhanced version of ResNet is utilized to capture spatial dependencies between mileposts; (2) Temporal Correlation Modeling, where an attention-driven LSTM network is employed to model the temporal dynamics of traffic flow; and (3) Weighted Feature Fusion, which integrates the spatial and temporal features to generate the final traffic flow predictions. The model incorporates three traffic flow patterns-real-time, daily, and weekly-allowing it to account for periodic traffic behavior and improve forecasting precision. Experiments conducted on real-world traffic datasets show that AST-Deep consistently outperforms nine baseline models, including traditional and machine learning approaches, by a significant margin in terms of forecasting accuracy. Specifically, the AST-Deep model achieves a 1 to 5 % improvement in mean absolute error and root mean square error over the best-performing baseline model as the prediction horizon increases. These results demonstrate the effectiveness of AST-Deep in capturing both spatial and temporal dependencies to provide more accurate and reliable short-term traffic flow predictions.</p>