Lithology identification method based on the PCA dimension reduction and improved PSO-XGBoost algorithm

Traditional lithology identification methods can not meet the actual needs due to the low identification accuracy and the poor generalization capability which were resulted from the imbalanced lithology distribution under the complicated geological conditions. With the rapid development of artificial intelligence, a new method for lithology identification has been provided through the application of machine learning. However, most existing studies typically focused on accuracy as the core optimization objective, which caused the models to overlook the issue of imbalanced lithological properties. In this study, the Macro F1 score is selected as the primary optimization objective, by using the PSO to optimize XGBoost hyperparameters and applying the PCA to reduce dimensionality, to effectively enhance the identification capability of the model. The model has excellent performance across various metrics in the lithology identification through the validation of test sets, with a Macro F1 score of 0.82, accuracy of 0.94, and an AUC value of 0.97. Notably, the achieved accuracies of lithology identification for both of the multi-sample mudstone and single-sample fine sandstone datasets are all 0.95. The excellent performance of this model in the lithology identification has effectively improved accuracy and efficiency for indentification of imbalanced lithology classes, has provided a more reliable solution for the intelligent lithology identification.