The increasing use of e-scooters in urban areas has raised safety concerns, necessitating research for effective safety interventions. This study analyzes three years of e-scooter crash data from the United Kingdom using statistical and machine learning methods to identify key factors influencing crash severity. We employed a random parameters logit model and investigated several machine learning algorithms, with XGBoost performing best. Analysis reveals that severe injuries are more likely in crashes involving senior riders, at night with lighting, and at T, staggered, or crossroad junctions. Further insights from the XGBoost-SHAP analysis and heterogeneity in means and variances of random parameters revealed nuanced patterns. While crashes involving female riders or crashes at give way or uncontrolled junctions typically have less severe outcomes, specific condition (young female riders or nighttime crashes at these junctions) intensify the risk of severe injuries. These insights advocate for tailored public policy adjustments and infrastructure enhancements to mitigate e-scooter risks, ensuring safer urban mobility for all demographics.