Predictive modeling of urban traffic congestion using optimized hybrid machine learning techniques

Urban traffic congestion is a growing challenge in rapidly expanding cities, adversely affecting mobility, environmental sustainability, and economic productivity. Traditional traffic management systems, often reactive and inflexible, struggle to accommodate the dynamic nature of modern urban road networks. This research develops preemptive solutions for detecting congestion on a continuous basis and in real-time, utilizing data-driven techniques. The primary goal is to construct and analyze machine learning (ML) models that can predict short-term traffic conditions in heterogeneous urban road networks. The case study applied a comparison framework by implementing five models, Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM), using a traffic dataset from Amman, Jordan. The dataset includes dimensions such as time, environment, and vehicle type. Furthermore, a bio-inspired algorithm called Ant Colony Optimization (ACO) is adopted for feature selection to improve model computational efficiency. Evaluation metrics include RMSE, MAE, and training time. LSTM outperformed all other models with an RMSE of 0.0195. Additionally, the ACO-optimized LSTM model still provided adequate accuracy while enabling a reduction in driven inputs, training time, and overall complexity. This strengthens the suitability of a hybrid approach for real-time applications in intelligent transportation systems (ITS). These results indicate that employing deep learning (DL) models along with metaheuristic optimization techniques improves strategies designed for urban traffic prediction.