Traffic congestion remains a critical challenge in urban mobility, requiring accurate and efficient
prediction models to enable intelligent transportation systems (ITS). Machine learning (ML) and deep
learning (DL) models such as Random Forests, XGBoost, Long Short-Term Memory (LSTM) networks,
and Graph Neural Networks (GNNs) have shown promising results in modeling nonlinear and dynamic
traffic patterns. However, their effectiveness largely depends on proper hyperparameter tuning, which
directly influences prediction accuracy and computational efficiency. This study investigates the
comparative performance of four hyperparameter optimization techniques—Grid Search, Random
Search, Bayesian Optimization, and Genetic Algorithms—for real-time traffic prediction. Using
benchmark datasets, each tuning method was applied to multiple models and evaluated in terms of Root
Mean Square Error (RMSE), Mean Absolute Error (MAE), training time, and latency. Experimental
results reveal that Bayesian Optimization consistently outperforms other approaches, achieving superior
accuracy with reduced computational cost, thereby offering the best balance between performance
and efficiency. The findings highlight the importance of efficient hyperparameter tuning in real-time
ITS applications and provide valuable insights for developing scalable, adaptive, and accurate traffic
prediction systems.
