A Novel SW-KMA-Bi-LSTM Approach for Improving Traffic Flow Prediction

Accurate traffic flow prediction is vital for reducing congestion and enhancing transportation efficiency.This issue becomes more challenging especially when implemented on wireless communication system as a significant part in Internet of Things era.This study proposes a novel model, Sliding Window-Komodo Mlipir Algorithm-Bidirectional Long Short-Term Memory (SW-KMA-Bi-LSTM), to enhance short-term traffic volume forecasting.The approach integrates the Sliding Window technique to capture temporal traffic patterns, Bi-LSTM for bidirectional sequence modelling, and the Komodo Mlipir Algorithm (KMA) for optimizing model parameters.Examinations under two benchmark traffic datasets obtained from the California Department of Transportation (CalTrans) Performance Measurement System, i.e., PEMS04 and PEMS08 have shown that the proposed SW-KMA-Bi-LSTM achieves the lowest RMSE of 18.1382 and MAE of 6.8404 on PEMS04, and RMSE of 13.6661 and MAE of 5.1887 on PEMS08 demonstrating superior prediction accuracy and outperforms the standard Bi-LSTM and Bi-LSTM models which optimized with various metaheuristic algorithms, such as GA, FOA, PSO, CWO, SOA, and AOOA.Although the model incurs higher computation time (365.64minutes for PEMS04 and 393.34 minutes for PEMS08), the accuracy improvements justify the runtime cost.These findings confirm that the proposed SW-KMA-Bi-LSTM is a reliable and effective model for supporting intelligent transportation systems through accurate traffic flow prediction.