<![CDATA[Skin cancer is a life-threatening dermatological disease, with malignant melanoma representing its most aggressive form. Early detection and accurate classification of skin lesions remain challenging despite advances in computer-aided diagnostic techniques. However, further comparative evaluation is still needed to determine whether a Voting Ensemble framework offers meaningful performance improvements over individual machine learning classifiers for malignant melanoma classification. This study aimed to design and evaluate an automated classification system using a Voting Ensemble framework and compare its performance with individual machine learning classifiers. The International Skin Imaging Collaboration (ISIC) dataset containing 10,000 dermoscopic images was preprocessed, and Principal Component Analysis (PCA) was applied for dimensionality reduction. Four machine learning models, namely Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), and Logistic Regression (LR), were evaluated alongside the proposed Voting Ensemble model. Experimental results showed that SVM achieved the highest classification accuracy (90.70%), outperforming the Voting Ensemble (89.50%), XGBoost (89.70%), RF (86.40%), and LR (81.60%). Although the proposed Voting Ensemble integrated the predictive strengths of multiple classifiers, it did not surpass the standalone SVM under the current experimental setting, indicating that SVM achieved the best classification performance under the selected dataset and experimental setting. These findings provide useful evidence for selecting appropriate machine learning models for automated melanoma screening and highlight the importance of rigorous comparative evaluation before adopting ensemble approaches in clinical decision-support systems. The results further suggest that ensemble learning does not necessarily outperform a carefully optimized standalone classifier under all experimental conditions.]]>
