A machine learning approach to quantify effects of geometric design features and traffic control devices on wrong-way driving incidents at partial cloverleaf interchange terminals

Abstract

This study addresses the issue of wrong-way driving (WWD) incidents at partial cloverleaf (parclo) interchange terminals in the United States. These incidents are a safety concern, often attributed to geometric design features and inadequate traffic control devices (TCDs). While previous research has acknowledged the significance of parclo interchanges as common initial entry points for WWD crashes, few studies have comprehensively quantified the impact of TCDs and design features on recurring WWD incidents.

In response, this study collected data from 75 parclo interchange terminals spanning 13 states. A subset of 28 ramp terminals exhibiting recurrent WWD incidents, amounting to 410 incidents during the data collection phase, received focused attention. Leveraging modern machine learning methodologies, two techniques: eXtreme Gradient Boosting (XGboost) and Lasso-logistic regression were applied to quantify the influence of distinct TCDs and design features on the probability of WWD incidents occurring.

The outcomes of this analysis revealed noteworthy results: the fitted XGboost model displayed an average accuracy of 80%, closely followed by the fitted Lasso-logistic regression model with an average accuracy of 78%. These models were subsequently employed to construct a practical network screening tool. This tool assists in the identification of potential WWD incident locations, predicated on the effects of TCDs and design characteristics.

The significance of this study lies in its potential to inform and guide state and local transportation agencies in enhancing the safety of parclo interchange terminals. By discerning the impacts of TCDs and design features, stakeholders can implement improvements that curtail the occurrence of WWD incidents, contributing to enhanced road safety and optimized traffic management.