TRAFFIC FLOW PREDICTION USING AN ATTCLX HYBRID MODEL

This study proposes an Attention-based CNN-LSTM-XGBoost (AttCLX) hybrid model to enhance short-term traffic flow prediction accuracy. The model integrates ARIMA for non-stationary data preprocessing, an Attention-based CNN-LSTM module for spatiotemporal feature extraction, and XGBoost for prediction refinement. Experiments using the PeMS dataset demonstrate that AttCLX outperforms benchmarks such as HA, ARIMA, SVR, LSTM, and DCRNN in both short-term (5-minute) and long-term (60-minute) predictions. Key metrics, including MAE and RMSE, show significant improvements (MAE: 13.69 for 5 minutes; 16.21 for 60 minutes). This research provides a robust solution for intelligent transportation systems to alleviate congestion and improve travel efficiency.