Prediction of Freight Truck Accident Severity and Identification of Risk Factors Using Machine Learning: Focusing on Gyeongsangnam-do

This study aims to predict the severity of freight truck traffic accidents in Gyeongsangnam-do, a major industrial and logistics hub, and to identify the structural factors that escalate accidents into high-risk outcomes (severe injury or fatality). To overcome the limitations of previous studies, such as data leakage and overfitting caused by using small sample sizes or post-accident variables, this study analyzed a total of 174,521 traffic accidents occurred from 2007 to 2024. We strictly excluded post-accident variables to prevent data leakage and used only pre-accident environmental variables. The SMOTE technique was applied to resolve class imbalance, and an XGBoost-based classification model was developed to predict high-risk accidents. Additionally, spatial analysis using QGIS was conducted to verify the regional distribution of risk factors.<br/> The results showed that freight truck accidents had a significantly higher rate of severe outcomes (56.7%) compared to passenger cars, proving their structural danger. Feature importance analysis identified 'freight truck operation' itself, along with 'collision with pedestrians/two-wheelers', 'major traffic violations (speeding)', and 'nighttime driving' as key triggers determining accident severity. Spatial analysis revealed high-risk accident hotspots concentrated in logistics routes connecting major industrial complexes in Changwon and Gimhae. This study contributes academically by presenting a realistic risk prediction model that addresses data limitations. Based on the findings, we suggest practical policy measures such as physical segregation of trucks and pedestrians and the implementation of total energy (speed) control systems.