Modelling service quality of two-wheelers at signalized intersections using Artificial Intelligence techniques

Service level assessment at signalized intersections is essential for effective traffic management, especially in urban areas with a high share of Motorized Two-Wheelers (MTWs). This study focuses on modelling the Motorized Two-Wheeler Level of Service (MLOS) using Artificial Intelligence (AI) techniques such as Multi-Gene Genetic Programming (MGGP) and Adaptive Neuro-Fuzzy Inference System (ANFIS). In this study Two independent AI-based models MGGP and ANFIS were constructed and evaluated, enabling a comparative assessment to identify the superior approach for accurately predicting MLOS scores. Dataset comprising intersection geometrics, traffic flow and operational variables were collected from 21 signalized intersections located in six mid-sized cities in India. Important parameters considered for model developments are peak hour volume, average control delay, turning radius, road surface condition and few others. MGGP was employed to derive interpretable mathematical expressions for MLOS prediction, while ANFIS utilized fuzzy logic integrated with neural networks to adaptively generate inference rules. Model performance was evaluated using R² and RMSE metrics, with ANFIS achieving an R² of 0.91 and RMSE of 0.36, outperforming MGGP which attained an R² of 0.88 and RMSE of 0.42. The results confirmed the suitability of both methods for capturing the nonlinear dynamics of heterogeneous traffic, with ANFIS offering superior predictive accuracy and MGGP contributing interpretability for engineering analysis. Fuzzy C-Means (FCM) clustering was employed to categorize MLOS scores into six distinct service levels, to provide realistic thresholds for traffic service quality assessment in two-wheeler dominated traffic flow.