Catastrophic causes of truck drivers’ crashes on Brazilian highways: Mixed method analyses and crash prediction using machine learning

Abstract

Traffic crashes represent a global challenge, especially in Brazil, where one-third of incidents on federal highways involve trucks, highlighting significant economic and safety risks for truck drivers and the community at large. This study focuses on understanding the specific causes of crashes involving trucks on Brazilian highways, using a decade of data from the Federal Highway Police to develop a predictive model aimed at accident prevention. It analyzes historical crash trends, selects attributes for prediction models, trains classifiers, evaluates predictions through confusion matrices, and enhances reliability via cross-validation techniques, aiming to develop an accident prevention tool. The analysis revealed a temporal pattern, with a slowdown in fatal incidents from 2013 to 2016, followed by an upward trend from 2017. MG-381 emerged as the deadliest highway, and single-lane roads were identified as more accident-prone, emphasizing the need for targeted preventive measures. Additionally, machine learning models achieved an accuracy of over 70 %, with XGBoost and LightGBM leading at 73 %, providing reliable insights for road safety interventions. In transportation engineering and road safety research, these findings highlight the importance of data-driven approaches to understand accident dynamics and design effective interventions to mitigate risks on highways, thereby contributing to increased road safety and social well-being.