HyperMinNet: A hypergraph-based framework incorporating high-order relationships for mineral prospectivity mapping

Abstract

Mineral Prospectivity Mapping (MPM) is a fundamental task in geosciences for identifying regions with high mineral potential. High-order geological associations are intrinsic to mineralization processes, yet have not been explicitly represented or effectively mined in existing studies, even though mineralization often arises from the joint influence of multiple geological factors interacting in complex ways. To address this problem, this study introduces HyperMinNet, a hypergraph-based framework for modeling and discovering high-order relationships in MPM. In HyperMinNet, hyperedges are constructed across both spatial and attribute domains to capture multi-factor geological associations. The Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm plays a central role in grouping geologically related units, thereby highlighting complex, spatially coherent associations beyond simple pairwise relationships. To further enhance the model’s ability to recognize informative geological patterns, an attention mechanism is employed to adaptively focus on critical associations, while a center loss function mitigates the influence of limited positive mineralization samples by enhancing the distinctiveness of learned representations. Experiments conducted in the Lhasa region demonstrate the effectiveness of the proposed framework, where HyperMinNet achieved an accuracy of 0.9042 and an Area Under the Curve (AUC) of 0.9511, confirming its strong potential in advancing mineral prospectivity mapping.