<![CDATA[Class imbalance in medical datasets, including prostate cancer, can affect the performance of machine learning models in detecting minority cases. This study compares three oversampling techniques - SMOTE, ADASYN, and Random Oversampling - to address data imbalance in prostate cancer classification. These techniques are applied to Random Forest (RF), Decision Tree (DT), and LightGBM (LGBM), which are evaluated using accuracy, precision, recall, F1-score, and ROC-AUC. In improving the reliability of the evaluation, K-Fold Cross Validation was used to reduce the risk of overfitting and ensure stable results. The findings show that oversampling techniques improve model performance compared to the baseline. Random Oversampling has the best performance for Random Forest with accuracy 0.85, recall 0.888, precision 0.873, F1-score 0.879, and ROC-AUC 0.838. SMOTE produced the highest Decision Tree performance with accuracy 0.80, recall 0.838, precision 0.843, F1-score 0.839, and ROC-AUC 0.788. ADASYN provided the most improvement for LightGBM, achieving accuracy 0.89, recall 0.919, precision 0.913, F1-score 0.913, and ROC-AUC 0.879. These results confirm that the oversampling method improves prostate cancer classification performance by tailoring the resampling technique to the model characteristics.]]>
Copyrights © 2025
