Deep Learning-Based Short-Term Traffic Flow Prediction for Smart Cities Using YOLO-NAS

The complexity of modern road networks and the ongoing rise in the number of automobiles have made urban traffic congestion a major issue. Accurate short-term traffic-flow prediction can support traffic monitoring and improve the understanding of changing traffic conditions. This paper presents a deep-learning-based framework for short-term city traffic-flow prediction using YOLO-NAS and CNN-Bi-LSTM. YOLO-NAS is employed to detect and classify vehicles from traffic surveillance videos and to extract important traffic features such as vehicle count, vehicle type, estimated speed, and traffic density. The extracted traffic information is organized as time-series data and used as input to a CNN-Bi-LSTM model. The CNN component extracts relevant local patterns from the traffic sequences, while the Bi-LSTM component learns temporal dependencies from historical traffic conditions to predict short-term future traffic flow. The prediction performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE). The proposed framework integrates real-time vehicle detection with temporal traffic-flow prediction and provides traffic information through a visualization interface. The approach demonstrates the potential of combining YOLO-NAS-based vehicle analysis with CNN-Bi-LSTM for intelligent traffic monitoring and short-term traffic-flow forecasting.