A semiautomatic multi criteria method for mineral resources classification

Abstract

Mineral resources are classified as measured, indicated, or inferred based on the geological confidence derived from estimation results and the quality and quantity of available data. In the mining industry, there are numerous accepted criteria for this classification. The selection and combination of these criteria, determined through arbitrary decision rules, depend on the characteristics of each project and are the responsibility of a qualified person. The problem with the current methodology is that the same data, using the same criteria, can yield different results depending on the person analyzing them. This contradicts the need for objectivity in the classification process, creates uncertainty in decision-making, and diminishes investor confidence. This article presents an innovative methodology for mineral resource classification using machine learning, which complements the work of the qualified person, reduces time costs, decreases subjectivity, and simplifies the reproducibility and auditability of the process, aligning with the fundamental principles of classification. The proposed methodology consists of three stages: first, classification criteria are selected, recording variables block by block. Then, blocks are grouped by similarity using the k-prototypes algorithm. Finally, the blocks are smoothed using a multilayer neural network to correct the "spotted dog" effect. This methodology is applied in two cases: a porphyry copper model and a high-sulfidation gold deposit in southern Peru. The results obtained reflect the fundamental principles of classification: transparency, objectivity, reproducibility, and auditability based on quantifiable principles. Additionally, the methodology ensures results consistent with the utilized characteristics, imposes no restriction on the acceptability of any type of input information, reduces time costs, and limits human intervention.