Hide/Show Apps

Assessment of discrimination of mafic rocks using trace element systematics with machine learning

Öztürk, Mehmet Sinan
Having an important role in the elucidation of the evolution of ancient oceans and related continental fragments, the determination of original tectonic settings of ancient igneous rocks is an essential part of the geodynamic inferences. Geochemical classification of mafic rocks is important for the tectono-magmatic discrimination of igneous rocks especially when geological information is insufficient as the link of the igneous rocks to their original tectonic setting had been erased due to large scale events. Starting from 1960s, first traditional methods (functions of elements or element ratios, bivariate and ternary diagrams of elements or element ratios), and then, recently, modern methods such as decision trees, support vector machines, sparse multinomial regression and random forest have been applied to develop tectono-magmatic discrimination methods. The purpose of this study is to assess new and better classification methods which are both statistically and geochemically rigorous using trace element systematics with decision tree learning, an effective machine learning method for classification. Dataset included a large number of samples well distributed through different tectonic settings (continental arcs, continental within-plates, mid-oceanic ridges, oceanic arcs, oceanic back-arc basins, oceanic islands and oceanic plateaus) as classes. Data is gathered from high quality articles which is known to follow accurate geochemical sampling procedures and have their samples analyzed in internationally accredited and trustworthy laboratories. Only element ratios have been used as features in order to increase the successful applicability of constructed decision trees to external datasets. With this study, successful decision trees with their alternatives are proposed for the tectono-magmatic discrimination between (1) subduction and non-subduction settings, (2) arc-related and back-arc-related settings within subduction settings, (3) oceanic arcs and continental arcs within arc-related settings, (4) oceanic and continental settings within subduction settings, (5) oceanic arcs and oceanic back-arcs within subduction-related oceanic settings, (6) mid-oceanic ridges + oceanic plateaus and oceanic islands + continental within-plates within non-subduction settings, (7) mid-oceanic ridges and oceanic plateaus within non-subduction settings and (8) oceanic islands and continental within-plates within non-subduction settings.