Separating volcanic rock groups: a novel method based on principal component analysis and a support vector machine

Separating volcanic rocks into groups based on field observations or geochemical data is important for understanding their origins. None of the conventional plotting methods for classification of different rock types are designed to do this. Isotope data may partly improve such classifications, but higher costs are involved. To address this issue, we applied principal component analysis (PCA) and a support vector machine (SVM) to geochemical data in order to separate volcanic rock samples into different groups. The first step used total alkalis–silica classification and Harker diagrams with an SVM to separate samples based on differences in rock type. We then used PCA and SVM together to verify the existence of a particular group of volcanic rocks and its relationship to geological processes. For the studied example, the results quantitatively verify the existence of a volcanic rock group and reveal a possible relationship between ore-forming processes and the recognized volcanic rocks. This study also demonstrates the feasibility of improvement based on some classical plots for quantitative classification of igneous rocks.