Driver risk-level identification incorporating personality traits, demographic characteristics, and driving behaviors

Human factors have a profound influence on the prevalence of vehicle crashes, particularly among high-risk drivers. This study aims to develop a driver risk-level identification method to effectively evaluate the safety of drivers and design training programs.

The personality traits of 50 drivers were quantitatively evaluated using the Symptom Checklist-90 scale, with parallel documentation of demographic information. A tri-level risk categorization (low, medium, and high) was established for drivers based on risky driving behaviors investigated via field tests. Correlations between driver personality traits, demographic characteristics, and risk levels were systematically analyzed. Driver risk-level identification models were developed utilizing four machine learning algorithms: support vector machines, classification and regression tree, eXtreme gradient boosting, and multi-layer perceptron, with the relevant personality traits and demographic characteristics serving as predictor inputs. Evaluation indicators were tested to determine the optimal method for driver risk-level identification.

Factors such as sex, obsessive-compulsive disorder, depression, anxiety, hostility, and paranoid ideation are significantly correlated with driver risk levels. Furthermore, all four models have a recall of 100%, demonstrating high efficacy in identifying high-risk drivers. Among the four methods, the multi-layer perceptron model achieves the highest overall accuracy (86.7%) and F1-score (87.0%), along with a precision of 90.5% and a recall of 86.7%.

Overall, personality traits and demographic characteristics play a crucial role in understanding driver risk levels, offering new insights to improve driver safety.