Speakers
Description
Abstract
Lost-time injuries (LTIs) remain a significant challenge across industries, particularly in high-risk sectors such as oil and gas, construction, mining, manufacturing, and transportation. Traditional safety management approaches are largely reactive, relying on post-incident analysis and compliance, which limits their effectiveness in preventing injuries. This study investigates whether machine learning can provide a more predictive framework for mitigating LTIs. Historical operational data, including LTI reports, worker demographics, task characteristics, shift duration, equipment status, and environmental conditions, were analyzed using five regression models: Linear Regression (LR), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Decision Tree (DT), and Gradient Boosting (GB). Model performance was assessed using the coefficient of determination (R²).
Results show that SVR outperformed all other models, achieving an R² of 99.40%, followed by GB (98.86%), DT (98.73%), and KNN (95.79%), while LR performed the weakest (78.74%). These findings confirm that LTIs are influenced by nonlinear and complex relationships that linear models fail to capture. Feature importance analysis identified prolonged shift durations, specialty maintenance tasks, and delays in equipment inspection as the strongest predictors of LTIs.
This research demonstrates the potential of machine learning to transform workplace safety management from reactive incident response to predictive prevention. By integrating these models into safety systems, industries can reduce operational downtime, improve compliance with safety regulations, and enhance field-level safety outcomes. Practical recommendations include cross-functional data integration, continuous retraining of predictive models, and adherence to industry safety standards. Overall, the study highlights machine learning as a scalable and data-driven solution for mitigating LTIs, particularly in sectors characterized by high operational risks.