This study advances pillar stress prediction and design optimisation in hardrock platinum mining on the Great Dyke of Zimbabwe using advanced machine learning (ML) techniques, addressing significant gaps in traditional methods. Utilising Gradient Boosting Machine (GBM), XGBoost, NGBoost, Random Forest, and AdaBoost, the research evaluated a dataset of 503 observed practical insitu pillars, incorporating key features such as Depth Below Surface (DBS), Actual Panel Width, and Actual Extraction Ratio (AER). GBM and XGBoost emerged as top performers, achieving R2 scores of 99.58% and 99.44%, respectively, with GBM exhibiting an MSE of 0.3094 and RMSE of 0.5563. NGBoost added value with predictive uncertainty, enhancing risk management frameworks. The study also highlights feature importance, emphasising DBS, AER, and Actual Pillar Area as critical predictors, ensuring robust and site-specific design solutions. Practical outcomes include a 15% reduction in material overdesign and a 20% improvement in identifying high-risk pillars, contributing to safer and more efficient operations. Integration with real-time monitoring systems enabled dynamic adjustments, reducing pillar failure risks by 30% under evolving conditions. This research, the first of its kind on the Great Dyke, demonstrates the transformative potential of ML in mining engineering, providing a framework for safer, economically viable, and sustainable operations. This study paves the way for leveraging ML to tackle complex geological and geotechnical challenges in global mining projects by addressing predictive accuracy and uncertainty.