<![CDATA[This study proposes a modified Directed Acyclic Graph Support Vector Machine (DAG-SVM) using a Rest-vs-Rest (RvR) strategy to address the multiclass classification problem in the Hepatitis C dataset from Kaggle, which contains four diagnostic categories with a highly imbalanced class distribution, with class sample sizes of 540, 24, 21, and 30, respectively. The aim of this study is to examine how hierarchical decision structures interact with extreme class imbalance in SVM-based multiclass classification. The method is implemented through three fixed hierarchical decision schemes {0,1} vs. {2,3}, {0,2} vs. {1,3}, and {0,3} vs. {1,2} which restructure the decision flow of conventional DAG-SVM. Experimental evaluation shows that although the proposed schemes achieve relatively high overall accuracy (0.91–0.93), the precision, recall, and F1-scores for minority classes remain extremely low. These findings offer a new empirical insight into how class imbalance propagates through the DAG hierarchy, leading to early elimination of minority classes, and highlight the need for imbalance-handling techniques such as resampling, cost-sensitive learning, or synthetic data generation. The contribution of this work lies in demonstrating the limitations of DAG-RvR under severe imbalance and providing a structured evaluation that can guide future improvements for more reliable multiclass recognition.]]>
Copyrights © 2025
