A Comparative Study of Machine Learning and Traditional Techniques for Grade Prediction and Grade-Tonnage Evaluation in a Small VMS Deposit

Abstract

Estimating grades in small-volume, high-grade volcanogenic massive sulfide (VMS) deposits can be difficult due to sharp changes in mineralization and limited data coverage around high-grade zones. This study compares ensemble machine learning models with interpolation and geostatistical methods to compare gold estimation and grade-tonnage results. Random Forest and Gradient Boosting were trained using drillhole composites and evaluated against Inverse Distance Weighting (IDW), Simple Kriging (SK), and Ordinary Kriging (OK). The trained models were applied across the block model to generate continuous grade predictions and support grade-tonnage calculations at multiple cutoff grades. The ensemble models showed lower RMSE and higher R2 values and captured grade patterns more efficiently than traditional methods. Grade-tonnage comparison indicated that IDW generated the highest contained gold equivalent at low cutoff grades, while OK and Gradient Boosting produced more consistent and geologically reasonable estimates. Overall, the results show that machine learning methods can complement traditional estimation techniques when combined with geological domain control and appropriate model tuning.