<![CDATA[Early detection of breast tumors is important to accurately distinguish benign and malignant lesions . This study aims to compare the performance of Random Forest and Naïve Bayes algorithms in the classification of breast ultrasound images based on texture and shape features. The dataset comes from BUSI which consists of two classes: benign 891 images and malignant 421 images, with images through the pre-processing stage, histogram feature extraction, GLCM, as well as morphological features (circularity and elongation). Feature rankings using Relief show that GLCM Homogeneity has the greatest contribution in distinguishing the two classes. Performance evaluation was carried out using K-Fold Cross Validation with variations of K=5, 10, 15, 20, and 25. The results showed that consistently placed Random Forest as the best-performing model. Random Forest achieved the highest accuracy at k-fold-5 at 74.18%, with stable AUC values at 0.771-0.777, sensitivity reaching 78.25%, and better specificity (62-63%) across the fold. In contrast, Naïve Bayes showed lower accuracy with a maximum value of 59% at k-fold-25, AUC in the range of 0.68, and low specificity (42-43%) despite the relatively high sensitivity. These findings confirm that across all k-fold validations, Random Forest remains the most balanced and reliable model for distinguishing benign and malignant in breast.]]>
Copyrights © 2026
