Abstract

Many local authorities use small-scale transport models to manage their transportation networks. These may assume drivers’ behaviour to be rational in choosing the fastest route, and thus that all drivers behave the same given an origin and destination, leading to simplified aggregate flow models, fitted to anonymous traffic flow measurements. Recent price falls in traffic sensors, data storage, and compute power now enable Data Science to empirically test such assumptions, by using per-driver data to infer route selection from sensor observations and compare with optimal route selection. A methodology is presented using per-driver data to analyse driver route choice behaviour in transportation networks. Traffic flows on multiple measurable routes for origin destination pairs are compared based on the length of each route. A driver rationality index is defined by considering the shortest physical route between an origin-destination pair. The proposed method is intended to aid calibration of parameters used in traffic assignment models e.g. weights in generalized cost formulations or dispersion within stochastic user equilibrium models. The method is demonstrated using raw sensor datasets collected through Bluetooth sensors in the area of Chesterfield, Derbyshire, UK. The results for this region show that routes with a significant difference in lengths of their paths have the majority (71%) of drivers using the optimal path but as the difference in length decreases, the probability of suboptimal route choice decreases (27%). The methodology can be used for extended research considering the impact on route choice of other factors including travel time and road specific conditions.