AI-Driven Safety Training Optimization for Industrial Workforces

Safety training remains a cornerstone of workplace injury prevention. Yet most organizations deliver standardized training without considering individual worker needs or predicting which training approaches will work best for different employees. This paper presents a machine learning framework that predicts training effectiveness and recommends personalized training strategies based on worker profiles. We constructed a synthetic dataset of 4,000 training records spanning six industries. The dataset captures worker characteristics, learning preferences, training delivery methods, and assessment outcomes. Five machine learning algorithms were evaluated for predicting whether training will achieve meaningful knowledge improvement. Gradient Boosting achieved the highest performance with an F1-score of 0.877 and ROC-AUC of 0.910, validated through 5-fold cross-validation. Feature analysis revealed that pre-test scores, delivery method, trainer quality, and attendance strongly influence training outcomes. Notably, hands-on practice and virtual reality training showed effectiveness rates of 86% and 79% respectively, substantially outperforming traditional classroom lectures at 47%, consistent with the literature-informed assumptions used to generate the dataset. The framework enables organizations to match workers with optimal training methods, prioritize high-risk employees for immediate training, and allocate training resources more efficiently.