Abstract: Specific computational tools assist geologists in identifying and classifying the lithology of rocks in oil well exploration, reducing costs, and enhancing operational efficiency. Machine learning methods integrate a vast amount of information, enabling efficient pattern recognition and accurate decision-making. This article categorizes the lithology of five oil wells in the Norwegian Sea, randomly dividing the data into a training set (70%) and a test set (30%). Using multivariate well log parameter data for training and validation, the application effectiveness of models such as Multilayer Perceptron (MLP), Decision Tree, Random Forest, and XGBoost is compared. The research results indicate that the XGBoost model outperforms others in terms of data generalization, achieving an accuracy of 95%. The Random Forest model follows with an accuracy of 94%. Meanwhile, Multilayer Perceptron (MLP) and Decision Tree models exhibit good robustness, with accuracies of 92% and 90%, respectively.