A semiautomatic multi criteria method for mineral resources classification.

This article focuses on an innovative methodology for classifying mineral resources—categorized as measured, indicated, or inferred—using machine learning to enhance objectivity, reproducibility, and efficiency in the mining industry. The approach involves three stages: selection of classification criteria by a qualified person (QP), clustering of blocks using the k-prototypes algorithm that handles both quantitative and qualitative data, and smoothing of classifications via a multilayer perceptron neural network to address spatial inconsistencies known as the "spotted dog" effect. Applied to a synthetic porphyry copper deposit and a real hydrothermal gold deposit in Peru, the methodology demonstrated transparent, auditable, and spatially coherent classifications consistent with geological confidence and estimation uncertainty. This semi-automatic process reduces subjective decision-making and time costs while allowing flexible incorporation of diverse criteria tailored to specific deposits.