Identification of potential suicide attempts by traffic accidents in the Czech Republic

Abstract

This study focuses on identifying hidden potential suicide attempts by traffic accidents (PSA-TA) in the Czech Republic by combining exploratory data analysis (EDA), classification algorithms (KNN, XGBoost), and interpretability (SHAP) to design a predictive model capable of distinguishing PSA-TA from ordinary fatal road traffic accidents (FRTA). The results of the analysis show that cases of suicidal behaviour exhibit specific characteristics. Typically, these involve collisions with a fixed obstacle outside residential areas, at times of low traffic, without the use of safety features, involving a single vehicle, and often involving male drivers. Based on these characteristics, the model identified 13 cases from 2024 that are likely to bear the hallmarks of intentional behaviour, even though they were officially recorded as ordinary FRTA. The results of the study confirm that advanced analytical tools can be used to detect hidden suicidal behaviour ex post, thereby contributing to more accurate statistics, forensic assessment of accidents, and the development of targeted preventive policies in the areas of traffic safety and mental health. We also outline ethical aspects and key limitations of administrative coding, including potential misclassification.