Comparison of manifold learning algorithms for identifying geochemical anomalies associated with copper mineralization

Le résumé fourni par la source

The Baiyin district, situated within the northern Qilian orogenic belt, hosts the largest concentration of copper mineral resources in Gansu Province, Northwestern China. Geochemical anomaly patterns are crucial indicators for mineral exploration in this region; however, they are frequently concealed within complex high-dimensional geochemical datasets. Moreover, the scarcity of labeled samples often restricts the effectiveness of supervised machine learning methods for accurate geochemical pattern recognition. This study utilizes unsupervised manifold learning algorithms, including Uniform Manifold Approximation and Projection (UMAP), t-Distributed Stochastic Neighbor Embedding (t-SNE), Isometric Mapping (Isomap), and Locally Linear Embedding (LLE) for identifying low-dimensional features closely associated with mineralization from high-dimensional geochemical datasets. The manifold learning algorithms were optimized by adjusting their key parameters through Receiver Operating Characteristic (ROC) test analysis to achieve optimal performance. The analytical results demonstrate that: (1) manifold learning algorithms exhibited superior performance over conventional factor analysis in accurately capturing complex nonlinear geochemical patterns; (2) The ROC curve and Area Under the Curve (AUC) values for the manifold learning algorithms were UMAP (0.711), t-SNE (0.693), Isomap (0.691), and LLE (0.652), indicating that the UMAP algorithm is the most suitable for identifying geochemical anomaly patterns in the study area; the prediction-area(P-A) analysis further confirmed the UMAP-derived anomalies with a relatively higher prediction efficiency; (3) manifold learning-driven high-probability zones exhibit significant spatial correlations with known mineral deposits, fault structures, and ore-bearing volcanic rock formations. These results highlight the superior capability of manifold learning techniques in extracting meaningful non-linear geochemical anomalies for further exploration of mineral resources.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.