Machine Learning–Based Modelling of Level of Service and Operating Speed on Multi-Lane Highways under Heterogeneous Traffic Conditions

Abstract: Accurate evaluation of highway performance is essential for planning, design, and operational analysis of multi-lane highways, particularly under heterogeneous traffic conditions commonly observed in developing countries. Conventional regression-based models often fail to capture the nonlinear relationships between traffic flow, roadway geometry, and performance measures such as operating speed and level of service (LOS). This study presents a machine learning-based framework for modelling operating speed and LOS on multi-lane highways using Artificial Neural Networks (ANN). Field data comprising traffic volume, percentage of heavy vehicles, and geometric characteristics were used to develop and validate the proposed models. The performance of ANN models was compared with conventional regression approaches using statistical indicators. Results demonstrate that ANN models provide superior predictive accuracy and better representation of complex traffic behaviour. The findings confirm the suitability of machine learning techniques for highway performance evaluation and provide practical insights for transportation planners and highway authorities.

Keywords: Level of service; Operating speed; multi-lane highways; Artificial neural networks; Heterogeneous traffic