Enhanced Lithology Classification in Well Log Data Using Ensemble Machine Learning Techniques

Reservoirs have a strategic role in areas where energy resources must be discovered and sustained. Information related to subsurface lithology is one of the important factors in formation evaluation and reservoir characterization. Conventional lithological interpretation has several drawbacks, such as long interpretation time and bias. These drawbacks are caused by geological heterogeneity, complex data sets, and large data volumes. Therefore, this research introduces an approach for lithology classification by integrating the Synthetic Minority Oversampling Technique (SMOTE) and machine learning models such as K-Nearest Neighbor (KNN), Random Forest (RF), Decision Tree (DT), and soft ensemble voting technique, which combines KNN, RF, and DT machine learning models. This research highlights that the KNN and Soft Ensemble Voting lithology classification models outperformed the other models, with nearly equal evaluation scores across all metrics. The KNN model consistently achieved the highest evaluation metrics, with an accuracy of 94.06%, precision of 94.13%, recall of 94.06%, and F1 Score of 93.57%. These results show that with appropriate preprocessing techniques and meticulous model tuning, KNN can independently achieve excellent accuracy, as demonstrated by the research results. The results of this research indicate the effectiveness of the proposed method in lithology classification, highlighting its potential to significantly improve conventional interpretation methods in the geophysical domain.