Generalized Additive Models to assess Truck Driver's Perceived LOS of Signalized Intersections

Modern freight transportation relies on trucks delivering millions of packages daily. There is limited research on urban infrastructure service quality, particularly signalized intersections for truck traffic use. Trucks experiences different service levels due to their low maneuverability, acceleration and deceleration. As India's economy grows rapidly in turn, resulting high truck demand in turn requiring truck service quality examination. This study collects traffic, geometric and conglomerative data from five Indian cities with 30 signalized intersections and 105 approaches. Truck drivers' satisfaction is collected at city outskirts where truck traffic is significant. Truck Level of Service (TLOS) is modelled using variables which have statistically significant correlation with perceived satisfaction score using Spearman correlation analysis. Generalized Additive Models (GAMs) use these variables as inputs. It is a intermediate modelling approach to black box machine learning algorithms and simple statistical models; which has a strong capability of handling the non-linearity in the perceived scores. In this study, six parameters affect TLOS, while some manuals use V/C ratio or control delay only for LOS estimation. Final GAM-TLOS model shows high accuracy with 0.92 R2. GAM-TLOS scores for service classes A-F obtained by linear interpretation of Likert Scale is valuable tool for Planners and designers.