This study presents the implementation of machine learning-based methodologies to address classification challenges in the definition of estimation domains, a key process in mineral resource estimation. To achieve this, databases from the Molejon deposit are utilised. The Molejon deposit is a low-sulfidation epithermal gold deposit located in the Molejón district, Panama. The primary mineralisation of interest consists of gold, with subordinate silver values and low concentrations of molybdenum and copper. Mineralisation is mainly distributed in quartz breccias, concentrated in quartz veins with a strong structural control; however, it can also be found near the surface in saprolites (Laudrum, 1995). Exploratory data analysis is a fundamental step in mineral resource estimation, culminating in the definition of estimation domains, which are subject to multiple sources of error (Emery, 2019). The definition of these domains is a time-consuming process with a significant manual component, relying on professional judgment to ensure their representativeness, robustness, and validity. In this context, studying alternative methodologies using machine learning tools becomes relevant, aiming to support informed decision-making by improving the accuracy and/or efficiency in estimation domain definition. This, in turn, enhances the representativeness of the resource model. Regarding the state-of-the-art in machine learning techniques, there is growing interest in their applications within the mining industry, particularly in resource estimation, geotechnics, and operational control. Specifically, for estimation domain definition, the work of Fustos (2017) is of particular interest, as it explores two geostatistical formalisms applied to clustering techniques to integrate geological knowledge into these algorithms (Romary et al, 2012). Additionally, the study conducted by Faraj (2021) proposes a workflow for defining estimation domains using hierarchical clustering, emphasizing geology, statistics, and spatial continuity in the process. For this study, the most suitable approach to support decision-making in estimation domain definition is considered to be unsupervised learning through non-hierarchical clustering techniques. The objective is to evaluate the performance of various algorithms within this category while ensuring the integration of geological knowledge into the process. Furthermore, an alternative methodology capable of incorporating the spatial dependency of regionalised variables into the clustering process is contrasted.