No abstract

No abstract

With the complicated geology of vein deposits, their irregular and extremely skewed grade distribution, and the confined nature of gold, there is a propensity to overestimate or underestimate the ore grade. As a result, numerous estimation approaches for mineral resources have been developed. It was investigated in this study by using five machine learning algorithms to estimate highly skewed gold data in the vein-type at the Quartz Ridge region, including Gaussian Process Regression (GPR), Support Vector Regression (SVR), Decision Tree Ensemble (DTE), Fully Connected Neural Network (FCNN), and K-Nearest Neighbors (K-NN). The accuracy of MLA is compared to that of geostatistical approaches, such as ordinary and indicator kriging. Significant improvements were made during data preprocessing and splitting, ensuring that MLA was estimated accurately. The data were preprocessed with two normalization methods (z-score and logarithmic) to enhance network training performance and minimize substantial differences in the dataset’s variable ranges on predictions. The samples were divided into two equal subsets using an integrated data segmentation approach based on the Marine Predators Algorithm (MPA). The ranking shows that the GPR with logarithmic normalization is the most efficient method for estimating gold grade, far outperforming kriging techniques. In this study, the key to producing a successful mineral estimate is more than just the technique. It also has to do with how the data are processed and split.

No abstract

No abstract