<p>This study addresses the challenges of limited data and class imbalance in predicting mineralization types in ore geology. Traditional methods like multi-class logistic regression often struggle with these issues, leading to overfitting and biased predictions. To overcome these limitations, we propose the Balanced Elastic Net Logistic Regression (BENLR) method and compare its performance with other machine learning techniques, including Support Vector Machine (SVM) and Over-Sampling approach. The BENLR method integrates Elastic Net regularization with class weighting to enhance model robustness and fairness. Elastic Net regularization combines L1 (lasso) and L2 (ridge) penalties, reducing overfitting and improving generalization. Class weighting ensures appropriate consideration of minority classes. The BENLR method, along with SVM and Over-Sampled Logistic Regression, was applied to a comprehensive dataset of geochemical compositions from various tin deposits in Malaysia, including elements such as Titanium (Ti), Vanadium (V), Manganese (Mn), Iron (Fe), Zirconium (Zr), Niobium (Nb), Antimony (Sb), Tungsten (W), Tantalum (Ta), Hafnium (Hf), and Uranium (U). The dataset comprised 28 samples categorized into four mineralization types: disseminated, hydrothermal vein, polymetallic, and pegmatite. The BENLR method demonstrated superior performance compared to traditional logistic regression, SVM, and Over-Sampled Logistic Regression. BENLR achieved an overall accuracy of 91% on the test data, with high precision and recall across most classes. Specifically, BENLR showed balanced performance, achieving high F1-scores for disseminated (1.00), hydrothermal vein (0.80), pegmatite (1.00), and polymetallic (0.67) classes. In comparison, Over-Sampled Logistic Regression achieved an accuracy of 82%, and SVM achieved an accuracy of 73%. BENLR's integration of Elastic Net regularization and class weighting resulted in more robust and fair predictions, effectively handling class imbalance and limited data. This study highlights the robustness of the BENLR method in handling limited and imbalanced data, providing more accurate predictions for classifying tin mineralization types. The findings underscore the potential of advanced machine learning techniques, specifically the BENLR method over SVM, and Over-Sampling technique, in mineral exploration and classification efforts. This research contributes to the field by offering more reliable and effective tools for predicting mineralization types, ultimately aiding in the identification of economically viable resources.</p>