<![CDATA[Breast cancer is one of the leading causes of death in women and relies heavily on early detection to improve the chances of recovery. The main challenges in machine learning-based early detection systems are class imbalance and the limitations of evaluation metrics that often rely solely on accuracy or ROC–AUC. In this context, the Matthews Correlation Coefficient (MCC) offers a more comprehensive assessment because it considers all elements of the confusion matrix. This study analyzes and compares the performance of four classification algorithms: Naive Bayes, Decision Tree, Random Forest, and SVM on the Breast Cancer Wisconsin (Diagnostic) dataset. The dataset was first divided using a stratified train–test split of 80:20 to maintain class proportions. Feature normalization was performed only on the training data to avoid data leakage, then applied to the test data using the same parameters. Furthermore, 5-fold cross-validation was performed on the training data for model evaluation and selection. The results show that SVM provides the best performance with an accuracy of 98.25%, a precision of 1.00, an F1-score of 0.9762, an AUC of 0.9971, and an MCC of 0.9630. Naive Bayes and Random Forest also show excellent performance with AUC values above 0.99 and an MCC of 0.9253, while Decision Tree has a lower performance. Confusion matrix and ROC curve analysis confirm the superiority of SVM in minimizing classification errors. These findings emphasize the importance of a multi-model approach and the use of MCC as a more representative evaluation metric in breast cancer early detection systems.]]>
Copyrights © 2026
