AI and Machine Learning Advances

The nature of industrial work places is hazardous in terms of complicated operations, dangerous material, and human factors. Predictive analytics based on the machine learning can provide new opportunities to enhance the accident prevention and safety risk management. The research assesses the usefulness of various machine learning algorithms to detect severity of accident during industrial accident by using an actual safety dataset of 439 accidents in the industrial environment based on three countries. The dataset contains the variables of industry sector, location of plant and workers, and the types of risk, which are the most critical, and the severity of accidents is handled as the multi-class classification issue. There were five supervised learning models (Logistic Regression, Random Forest, Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Artificial Neural Networks (ANN)) that were used and tested based on an 80/20 train-test split. Accuracy, precision, recall, F1-score and confusion matrix were used to evaluate the performance of the model. Findings indicate that ensemble algorithms, especially XGBoost and Random Forest, are good predictors. The analysis of feature importance reveals that the possible level of accidents, time-related factors, the location of a plant, and the important risk categories are considered to be important predictors of the accident severity.