Supervised geochemical anomaly detection by pattern recognition

摘要: Geochemical anomaly detection is an important issue in mineral exploration. The availability of a training dataset consisting labeled geochemical samples background and classes enables us to define supervised pattern recognition framework for detection. Therefore, various classification feature selection algorithms can be utilized build predictive model classify the unseen into pre-defined classes. In this study, some state-of-art were Kuh Panj porphyry-Cu district. Filter, wrapper embedded mode used remove redundant irrelevant elements from procedure. Subsequently, AdaBoost (ADB), support vector machine (SVM) Random Forest (RF) trained with borehole surface rock drilled parts study area create classified map depicting anomalous areas undrilled Results show that could play role increasing accuracy generalization ability classifiers used. Wrapper subset method combined genetic algorithm (GA) search resulted best performance area. Applied outperform Gaussian linear discriminant analysis (GLDA) provide more accurate, robust reliable results. Among applied methods, ADB achieved leave-one-out cross-validation (LOO) error rate 0.06. Meanwhile, comparison using another one created via concentration–area fractal indicated advantage former terms detecting high-promising prospective target region.