Comparison of geochemical data from Western Anatolian and cycladic syn-extensional granitoids using machine learning

Binary variation diagrams are traditionally used to reveal differences and similarities by utilizing geochemical data obtained from rock samples. However, when this approach is applied to data sets with many variables, several limitations arise. In recent years, machine learning methods have emerged as an alternative for analyzing multivariate data. These methods can be applied to explain geochemical data through supervised and unsupervised learning approaches. In this study, geochemical data from Miocene-aged granitoids exposed in Western Anatolia and the Cyclades—previously published by various researchers—have been evaluated. The method was applied to a dataset formed from analytical results of these granitoids, and particularly the geochemical characteristics of major, trace, and rareearth elements of the granitoids from Western Anatolia and the Cyclades with in the same age range were compared to reveal their similarities and differences.The most suitable methods for classification, component effects, etc., in supervised and unsupervised learning algorithms were identified, and comparisons were also made with granitoids around the world that exhibit similar geochemical characteristics. In the supervised learning approach, when the characteristic elements selected by the decisiontree algorithmare evaluated together with the selected training data and the Western Anatolia and the Cyclades data and binary diagrams, it has been observed that they show correlations with certain elements, but occasionally deviate from their respective fields. This demonstrates a significantad vantage of machine learning—its ability to evaluate multiple elements simultaneously. Therefore, binary diagrams are not always sufficient when compared with training data; in contrast, machine learning provides more reliable and comprehensive results. The results generated by the supervised learning method are highly meaningful and robust. Considering the characteristics of thetraining data correlated with the Western Anatolia and Cyclades units, it is evident that these units share similar geochemical, petrogenetic, and geodynamic properties. It has also been observed that there is a direct tectonic connection between the Aegean region and regions such as the South China Block, North China Craton, and Tarim Craton (North west China), and that the granitoids in these areas show similar evolutionary characteristics in terms of geodynamic processes. Although the unsupervised learning method can create internally meaningful clusters, these clusters often fall short in terms of direct overlap. Based on all geochemical evaluations, it was investigated whether granitoids with similar geochemical characteristics worldwide were formed in similar geodynamic settings. When all data are considered, it is recommended that supervised learning methods can be preferred over unsupervised ones for classifying granitoids and revealing their geochemical properties.