Accident analysis in traffic for the province in Malatya

This study conducts a data-driven analysis to determine the severity levels of traffic accidents occurring within the urban road transportation system of Malatya province. The main objective is to reveal the cause–effect relationships of traffic accidents and to predict their possible outcomes using machine learning methods. Within this scope, large-scale accident data obtained from the General Directorate of Security (EGM) were utilized, and data preprocessing steps including missing observation removal, feature scaling, and transformation of categorical variables were applied. Subsequently, the Synthetic Minority Oversampling Technique (SMOTE) was employed to eliminate data imbalance, and the dataset was divided into training (80%) and testing (20%) sets. During the modeling stage, traffic accident severity was predicted using Logistic Regression, Decision Tree, Random Forest, Support Vector Classification (SVC), AdaBoost, and XGBoost algorithms. Model performance was evaluated using accuracy, precision, recall, F1-score, and Cohen's Kappa coefficient. The results indicate that the Logistic Regression model demonstrated a more balanced and generalizable performance, while XGBoost was more successful in capturing complex patterns in the dataset, and the Decision Tree model exhibited a tendency toward overfitting despite its high fitting capability. The analysis results reveal that variables such as driver age and behavior, road characteristics, pedestrian behavior, and the year in which the accident occurred have significant effects on accident severity. This study constitutes an important example for shaping urban traffic safety policies through data-driven approaches and provides guiding insights for traffic safety planning in cities with similar socio-economic and structural characteristics.