Enhancing urban traffic congestion prediction through efficientnet and optimized ensemble learning models

Traffic congestion is a growing concern in urban areas, impacting both infrastructure and quality of life. Predicting traffic flow accurately is essential for proactive traffic management. This research proposes a novel hybrid deep learning approach for traffic flow prediction, leveraging an ensemble of Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Bidirectional Gated Recurrent Unit (BiGRU) models. Features are extracted using EfficientNet, with hyperparameter tuning optimized by the Eurygasters Optimization Algorithm (EOA). The ensemble model's performance is further enhanced using Tournament-Selected Glowworm Swarm Optimization (TSGSO), improving prediction accuracy. Fuzzy logic is applied to classify traffic flow severity into low, medium, and high congestion levels. The proposed method is validated using traffic data collected from four urban intersections, demonstrating improved accuracy compared to existing models. Experimental results show that the ensemble model, combined with TSGSO optimization, outperforms traditional machine learning techniques in prediction accuracy while maintaining reasonable processing time. This research provides a scalable and efficient solution for real-time traffic management, contributing to smarter cities and reduced congestion.