Abstract:

This study suggests that data mining methods can be helpful in preventing occupational accidents in the textile industry. Within the scope of the study, 89.963 occupational accident data that occurred in the textile sector between the years 2019-2021 were examined and the number of samples was reduced to 11.710 with the data preprocessing study. In estimating accidental injury types, model selection map was taken as reference and SVM, Extra Trees, Random Forest, Gradient Boosting and XGBoost algorithms were chosen. Models were compared using the macro F-score performance metric. The estimation performance of models has increased with data balancing and parameter optimization methods. XGBoost algorithm performed better than other algorithms with 70% prediction success. The SVM (69%) and Extra Trees (68%) have been among the algorithms that correctly interpreted the data set by reaching high macro F-score values. It has been seen that the features that have the most effect on the estimation result are cause of accident, material agent, sub-sector, and company size, respectively.

Keywords: