A hybrid Machine learning and statistical modeling approach for analyzing the crash severity of mobility scooter users considering temporal instability

Abstract

One of the main objectives in improving the quality of life for individuals with disabilities, especially those experiencing mobility issues such as the elderly, is to enhance their day-to-day mobility. Enabling easy mobility contributes to their independence and access to better healthcare, leading to improvements in both physical and mental well-being. Mobility Scooters have become increasingly popular in recent years as a means of facilitating mobility, yet traffic safety issues such as crash severity have not been adequately investigated in the literature. This study addresses this knowledge gap by employing a hybrid method that combines a machine learning approach using the eXtreme Gradient Boosting (XGBoost) algorithm with Shapley Additive exPlanations (SHAP) and an advanced statistical model called Random Parameters Binary Logit accounting for heterogeneity in means and variances. Analyzing the United Kingdom mobility scooter crash data from 2018 to 2022, the study examined temporal instability using a likelihood ratio test. The results revealed that there was instability over the three distinct periods of time based on the coronavirus (COVID) pandemic, namely, pre-COVID, during COVID, and post-COVID. Moreover, the results revealed that mobility scooter crashes occurring at a give-way or uncontrolled junctions has a random effect on the severity, while factors such as mobility scooter riders aged over 80, rear-end and sideswipe crashes, and crashes during winter months increase the risk of severe injuries. Conversely, mobility scooter riders involved in crashes while riding on the footway are less likely to experience severe injuries. These findings offer valuable insights for enhancing road safety measures that can be utilized to effectively reduce the crash severity of mobility scooter riders.