Assessing Travel Time Prediction Models for Mixed Traffic on a Two-Lane Highway: A Case Study of the Dhankhola-Bhaluwang (H01) Road Section

Abstract

Over the years, substantial efforts have been dedicated to improving travel time prediction for corridors, using a multitude of variables. However, predicting travel time in corridors remains an inherently challenging task due to the intricate interplay of numerous factors, which are often difficult to comprehensively collect. This challenge is particularly pronounced in undivided roads, where corridor access is unrestricted, leading to a heightened presence and influence of various influencing factors. The present study is focused on the development of a travel time prediction model for the Dhankhola-Bhaluwang road section, which is a two-lane, two -way undivided highway, for both directions. Using an extensive analysis of 72-hour datasets on vehicle travel times, sourced from traffic volume counts and speed surveys, this study evaluates the effectiveness of travel time prediction models, taking into account through traffic, opposing traffic, and the proportions of heavy vehicles in through traffic. Evaluation metrics derived from Random Forest Regression consistently outperform those of other regression models for both directions. Subsequently, Support Vector Regression, Decision Tree Regression, LASSO Regression, and Multiple Linear Regression follow in effectiveness.