Application of machine learning methods to classify quartzites by chemical composition: the influence of trace elements and geochemical identification

Research subject. Quartzites from Different Sites of the East Sayan Quartz-Bearing Region. Aim. The aim of this study is to apply machine learning methods to effectively classify quartzite samples by their chemical composition, including the identification of key trace elements such as manganese and the detection of geochemical differences between samples. Materials and methods. The study used chemical analysis data from 776 quartzite samples, which were interpreted using machine learning methods. Standard data preprocessing techniques such as normalization were applied as well as data augmentation using SMOTE to solve the class imbalance problem. As a result, the CatBoost algorithm was selected, which showed high classification accuracy. Results. Cross-validation results showed that the CatBoost algorithm achieved classification accuracy of up to 97 %. The importance of features indicates that manganese is a key element in the classification of samples, while elements such as aluminum and potassium have a supporting effect. The analysis of the classification by color of quartzites with an accuracy of 0.94 was also successfully carried out. Conclusions. The study demonstrates the effectiveness of applying machine learning methods to the analysis of the chemical composition of quartzites, providing new opportunities for geochemical and archaeological research.