INJURY SEVERITY MODELING IN MINING INDUSTRY USING WORKERS' COMPENSATION DATA AND DATA ANALYTICS

Abstract

Early implementation of risk analysis can assist in preventing industrial accidents and their detrimental repercussions including fatalities, physical impairments, and monetary loss. Workers' risk of being involved in workplace accidents and occupational injuries can be managed if the dangers causing them are known beforehand. The type of risk associated with an injury in the mining industry can be identified through the analysis of workers’ compensation (WC) data. The data also talks about the extent of an accident or injury severity at a mine site. Variables like age group, gender, injury cause, nature of injury, part of body injured, and type of mine were considered for risk factor analysis. The significance of these variables on the compensation claim type was analyzed to understand the injury severity. The claim type was classified into medical only (0) and lost time injuries (1). The relationship between the risk factors and claim type from the WC data was evaluated through Chi-squared and post-hoc tests. These tests help in identifying the statistical association between risk factors and injury severity. The odds ratios were calculated to understand the degree of association in a multivariate setting using the logistic regression weights. The study was carried out using the WC data from the mining industry in Alaska. Results from the statistical analysis show that age group, injury cause, nature of injury, part of body injured are statistically significant risk factors, whereas gender and mine type aren’t significant with injury severity. Once the statistically significant risk factors and measures of impact were analyzed, the machine learning models were applied. The categorical variables were treated using one-hot encoding before applying ML models. The data was divided into the training and testing set. Class balancing and data augmentation techniques like synthetic minority oversampling technique (SMOTE) and adaptive synthetic sampling (ADASYN) were used to deal with the class imbalance problem. Tree- based algorithms like decision tree, random forest, and gradient boosting with a few other algorithms like logistic regression and support vector machine were tested to understand the ML performance. The gradient boosting algorithm with ADASYN (data augmentation technique) proved to be the best machine learning model to predict injury severity. The results of statistical analysis and machine learning algorithms can help the mine personnel comprehend risk factors by taking preventative actions at the mine site.