Short Term Traffic State Prediction via Hyperparameter Optimization Based Classifiers.

Area of Science:

• Intelligent Transportation Systems (ITS)
• Machine Learning Applications in Transportation

Background:

• Short-term traffic state prediction is vital for advanced traveler information systems (ATIS).
• Existing models struggle with traffic's dynamic and stochastic nature, often leading to overfitting.
• Current approaches focus on speed, density, or volume data, with questionable accuracy.

Purpose of the Study:

• To address overfitting in short-term traffic prediction using state-of-the-art models.
• To enhance traffic state prediction accuracy through hyperparameter optimization.
• To evaluate machine learning classifiers for traffic state modeling.

Main Methods:

• Utilized hyperparameter optimization via random sweep.
• Applied machine learning classifiers: Local Deep Support Vector Machine (LD-SVM), Decision Jungles (DJ), Multi-Layers Perceptron (MLP), and CN2 rule induction.
• Evaluated traffic states using Level of Service (LOS) and if-then rules.

Main Results:

• Hyperparameter optimization yielded significant improvements, averaging over 95% performance enhancement.
• Decision Jungles (DJ) and LD-SVM achieved high accuracies of 0.982 and 0.975, respectively.
• Decision Jungles demonstrated superior robustness and performance compared to other methods.

Conclusions:

• Hyperparameter optimization is effective for improving short-term traffic prediction accuracy.
• Machine learning models, particularly Decision Jungles, offer robust solutions for traffic state prediction.
• The study highlights the potential of advanced ML techniques in intelligent transportation systems.