Application of Machine Learning Techniques in Delay Modelling for Heterogenous Signalized Intersections

Abstract

Delay can be defined as the amount of extra time spent by a vehicle at an intersection. It is important to estimate vehicular delay as accurately as possible so that the efficiency of the intersection can be determined in terms of Level of Service (LOS). In general, researchers tend to opt either US-based HCM or UK based Webster's model to estimate the delay for Indian traffic conditions. As these models are designed for homogeneous and lane-based behavior, they tend to give erroneous results when applied for Indian road conditions. So, a model is developed to estimate the delay under heterogenous and no lane discipline conditions. As deep learning techniques provide high accuracy prediction, this paper is an attempt to estimate delay using Support Vector Machine (SVM) and Artificial Neural Network (ANN). In this study, video graphic survey is conducted to collect required parameters such as volume, signal timing, geometric features such as the number of lanes and approaches, the proportion of vehicles crossing in the green interval, platoon ratio and field delay is estimated by using HCM 2016 procedure. Two predictive models are formulated and compared to find out the best model which predicts the delay accurately. From the developed two models it can be concluded that ANN is the best fit model to estimate the delay at a signalized intersection as the R^2 (0.88) value is more when compared with R^2 (0.59) obtained from SVM Model.