Work Place Safety: Machine Learning Techniques for Assessing Workplace Incident Severity

Occupational injury severity prediction is paramount across all industries, necessitating the adoption of advanced methodologies to enhance predictive analysis. This study leverages Machine Learning (ML) techniques to build a robust and effective predictive model for predicting the severity of occupational injuries. Analyzing a comprehensive dataset of occupational injury records from Occupational Safety and Health Administration (OSHA), this research explores the efficacy of seven distinct ML algorithms AdaBoostClassifier (ABC), K-Nearest Neighbors (KNN), Decision Tree Classifier (DTC), Extreme Learning Machine (ELM), Support Vector Classifier (SVC), Multi-Layer Perceptron (MLP), and Stacked Model (SM). Among these, ABC emerges as the standout performer, demonstrating superior accuracy and F1-score, Specifically 92.5% and 99.3% accuracy in predicting hospitalization and amputation severity, respectively, thus underscoring the potential of ensemble learning in refining occupational injury prediction models. Permutation feature importance analysis reveals that the nature of the injury, the body part affected, and the source are pivotal factors in the dataset shaping injury severity predictions. This study empowers stakeholders to devise informed corrective and preventive strategies for mitigating workplace injuries. ABC has shown robust performance in developing injury prevention frameworks.