Akademik Veri Yönetim Sistemi
Natural Resources Research, cilt.35, sa.2, ss.729-750, 2026 (SCI-Expanded, Scopus)
Predicting the value of mineral properties (also referred to as mineral assets) is challenging due to external market dynamics such as commodity boom–bust cycles and price fluctuations, market volatility and global financial trends, as well as project-specific factors such as the development stage of an asset, data availability or location-based risk factors that significantly affect the value of mineral assets. This study utilized comparable transaction data of gold properties to establish a real and objective basis for a data-driven valuation approach. A range of machine learning algorithms was employed including linear (LR), lasso (LaR), ridge (RR) and polynomial (PR) regressions, support vector machines (SVM), decision trees (DT), random forest (RF), artificial neural networks (ANN) and extreme gradient boosting (XGB) to predict gold property values. These methods were applied to evaluate value-related factors, quantify their influence on asset value and compare model performances. The results indicate that complex models such as tree-based methods and artificial neural networks outperformed simpler models, as expected. In the non-producing assets dataset, RF achieved the highest coefficient of determination (R2) of 0.95, followed closely by ANN, XGB and SVM with R2of 0.95, 0.94 and 0.94, respectively. In contrast, PR and LR yielded the lowest performance with R2 of 0.71 and 0.77, respectively. Similarly, in the producing assets dataset, XGB, ANN and RF outperformed the other models with R2 of 0.95, 0.89 and 0.88, respectively, whereas LR, LaR and RR showed the weakest performance with R2 of 0.72, 0.78 and 0.79, respectively. Based on root mean squared errors (RMSEs), XGB, RF and ANN demonstrated high performance for both datasets. The RMSEs of these models for producing mines were 194.9, 507.3 and 483.4, while for non-producing mines the RMSEs were 68.2, 41.4 and 69.3, respectively. The paper demonstrates that advanced machine learning techniques such as RF, XGB and ANN can provide accurate and objective predictions of gold property values. The key value drivers identified by the models were gold equivalent and tonnage as the most influential, while open-pit operations tend to contribute more to value rather than underground methods. These findings offer a data-driven and interpretable framework for understanding the determinants of mineral asset valuation.