Analyzing predictive factors influencing helmet-wearing behavior among E-bike riders

Abstract

E-bikes revolutionize urban transport by offering an affordable and cost-effective alternative. However, poor helmet-wearing habits pose significant safety concerns, requiring effective interventions to mitigate head injuries and fatalities. This study investigated key factors predicting helmet-wearing behavior among e-bike riders in Guangdong Province, China, based on data from 14,762 survey valid responses. Logistic regression and three machine learning models: Random Forest (RF), XGBoost, and Support Vector Machine (SVM) were applied to predict helmet use and identify associated risk factors, with the RF model demonstrating superior predictive performance, achieving 91 % accuracy and 97 % Area Under the Curve (AUC). Using SHAP analysis, the study interpreted the influence of each factor based on the RF model revealing gender, riding experience, age group, average monthly income, policy management, and safety activity effectiveness as significant predictors of helmet-wearing behavior. For instance, SHAP waterfall plots for the first dataset showed that being male and receiving safety education through new media (e.g., WeChat, Weibo) raised the likelihood of non-helmet use by +0.03 and + 0.01, respectively. SHAP dependence plots further uncovered complex non-linear correlations, highlighting those males, inexperienced riders, and younger riders (under 18, 18–25, and 26–35) were less likely to wear helmets. Heatmap analysis indicated that diverse safety education methods combined with enriched content were strongly associated with increased helmet-wearing. Findings suggest that targeted safety campaigns, improved policy management, and stricter enforcement, supported by regular monitoring and evaluation, are essential to reduce non-helmet use and improve e-bike rider safety. Future research should use longitudinal studies and e-bike crash data to assess how safety education and policy interventions affect helmet-wearing patterns, crash rates, and injury severity over time.