Multi-Class Prediction of Occupational Accident Types Using XGBoost in Pipeline Construction

Abstract: In this study, the XGBoost algorithm was chosen to predict accident types using data from incidents that occurred in pipeline construction projects, and a multiclass classification approach was presented. A model was trained on 1,184 real accident records obtained from pipeline construction projects. This data includes information such as the time of the accident and the type of treatment required. Classification results showed an overall accuracy of 83%, a macro-average F1 score of 0.82, and a weighted F1 score of 0.81. Some classes showed high recall rates, while recall was low in categories with limited data. The results demonstrate that XGBoost can be a suitable model for predicting accident types using historical incident datasets and can contribute to risk analysis in construction projects through artificial intelligence.