Transforming workplace safety through leveraging predictive analytics and explainable AI in steel industries

Abstract

Workers in the steel manufacturing plants often confront perilous working conditions characterized by limited visibility, potential hazards from heavy machinery interacting with pedestrian staff, and the dangerous dynamicity of manufacturing processes. Such environments involve repetitive tasks, extreme temperatures, high noise levels, and challenging surroundings, fostering situational and behavioral risks that escalate the likelihood of accidents leading to injuries, illnesses, or fatalities. Therefore, it is imperative to scrutinize safety incidents within the steel industries to mitigate risks and enhance safety measures proactively. This study employs Machine Learning (ML) to develop predictive models using a dataset comprising 3600 workplace incidents reported from year 2018 to 2022 from three integrated steel manufacturing plants in India. The aim is to identify conditions indicative of unsafe events or situations based on different ML models. Five ML models were compared viz. Random Forest, Gradient Boosting, Support Vector Machine, Decision Tree, and K-Nearest Neighbor. Random Forest emerged as the most effective, achieving 86.52% accuracy and 100% AUC score in three-class classification. The classification of accident types provides valuable insights into potential risks, enabling proactive measures to prevent future incidents. Through the appropriate identification of conditions that lead to specific types of accidents, this research offers a data-driven approach to enhance workplace safety protocols. Furthermore, this study contributes significantly to Explainable AI (XAI), such as Local Interpretable Model-Agnostic Explanations (LIME), particularly in enhancing workplace safety approaches in the Indian steel industry.