Evaluating StackingC and ensemble models for enhanced lithological classification in geological mapping

Lithological classification is a crucial aspect of mineral exploration, providing insights into rock and mineral types in a given area. Conventional methods for lithological classification can be limited in terms of coverage, accuracy, and efficiency, often experiencing significant time and cost. Machine learning techniques have demonstrated considerable potential in improving the efficiency and accuracy of this process. In this study, the effectiveness of ensemble learning models, including boosting, stacking, and bagging, was compared to logistic regression (LR) and support vector machines (SVM) as baseline models for predicting lithological classes using geochemical and geological data. Notably, the stackingC model, a novel stacking variant, stood out as the best-performing model. It achieved remarkable Cohen's Kappa and Matthews Correlation Coefficient (MCC) scores of 97.10% and 93.70%, respectively. The Bagged Decision Trees and Adaboost models also demonstrated strong performance, with a kappa score of 97.10% and an MCC of 92.80%. In contrast, the LR model underperformed, scoring 37.70% in kappa and 43% in MCC. These results emphasize the potential of ensemble learning models for lithological classification, mainly when dealing with complex, nonlinear relationships between input variables and output labels. Such models hold promise for improving accuracy and generalization in mineral exploration. •Traditional methods limit lithological accuracy, prompting the need for improvement.•Ensemble learning enhances lithological classification in mineral exploration.•Ensemble learning enhances mineral exploration with nonlinear relationships.