Using an Interpretable Machine Learning Framework to Understand the Relationship of Mobility and Reliability Indices on Truck Drivers’ Route Choices

Abstract

This paper explores the relationships between travel time index and planning time index, two proven indices used to measure real-time congestion and travel time reliability, and their impacts on truck drivers’ route choices through training of a predictive model based on a machine learning algorithm—eXtreme Gradient Boost (XGBoost). Moreover, this study adopts an interpretable machine learning framework called SHapley Additive ExPlanation (SHAP) in the predictive model to reveal the insights usually hidden inside the machine learning “black box.” The predictive model is trained through a truck trajectory dataset provided by the Maryland Department of Transportation State Highway Administration and INRIX. The classical logistic regression model is adopted as the baseline model. The results show the XGBoost model can better handle nonlinearity and provide more reliable predictions. Through the SHAP framework, the results indicate that mobility and reliability indices and total trip time nonlinearly influence route choices. Truck drivers are more sensitive to real-time congestion information and reliability information when the differences in mobility and reliability indices on candidate routes reach certain thresholds. Moreover, the interaction study on trip time and mobility index shows that truck drivers are more sensitive to real-time congestion information if the candidate routes’ travel time is a larger portion of the total trip time.