A Systematic Review of Unsupervised Geochemical Anomaly Detection Methods

Geochemical anomaly detection is a fundamental task in mineral exploration, as it supports early-stage targeting and risk reduction. High-dimensional geochemical datasets contain complex nonlinear relationships that limit the effectiveness of traditional statistical methods. In recent years, unsupervised and semi-unsupervised machine learning approaches have become dominant tools for anomaly identification. This study presents a critical review of eleven representative methods applied to multivariate geochemical anomaly detection. The reviewed approaches include deep learning models, ensemble algorithms, clustering techniques, and geometry-based outlier detectors. Each method is evaluated based on learning strategy, data structure, robustness, interpretability, and exploration relevance. The analysis shows that deep autoencoder-based models and hybrid frameworks outperform conventional methods in handling high dimensionality and noise. Geological constraints improve spatial coherence and model credibility. Parameter-free and ensemble approaches provide stable performance in complex datasets. Despite these advances, challenges remain in model interpretability, data dependency, and transferability across geological settings. This review clarifies methodological trends and provides guidance for selecting suitable anomaly detection strategies in mineral exploration.

کلیدواژه ها:

Anomaly detection ، Unsupervised learning ، Mineral exploration ، Deep autoencoder-based models