Analysis of clustering methods for geochemical anomaly identification through weighted sample catchment basins

The cruciality of geochemical exploration in discovering new mineral deposits demands the use of high-performance computational techniques to recognize geochemical anomaly patterns associated with mineralization. High dimensionality and complexity of geochemical datasets lead to use of brand-new methods including machine learning tools to get more efficient and accurate outcome. In this regard, advanced clustering methods have gained significant popularity over the years in the field of mineral exploration. This study presents an integrated mineral exploration approach using multiple clustering algorithms to identify Cu mineralization potential in the Kariz-Now district, Razavi Khorasan, northeastern Iran. The methodology uses faults within the study area to weight the catchment basins and brings grade characteristics of role-playing elements along with fault weights to highlight the potential areas. Three clustering techniques including K-Means (KM), Fuzzy C-Means (FCM), and DBSCAN were applied to the variables attached to their corresponding sample catchment basins to predict potential areas. The results were validated by success rate curves of the selected methods and the area under curve was calculated. The results reveal that DBSCAN has higher area under curve (AUC) than KM and FCM. Furthermore, there is a strong correlation between predicted anomalies and Paleogene volcanic and sedimentary rocks, particularly in areas with high fault density.