Software Defect Prediction is crucial to ensure software quality. However, highdimensional data presents significant challenges in predictive modelling, especially identifying the most relevant features to improve model performance. Therefore, efforts are needed to address these issues, and one is to apply feature selection methods. This study introduces a new approach by applying the Copeland ranking method, which aggregates feature weights from multi-wrapper methods, including Recursive Feature Elimination (RFE), Boruta, and Custom Grid Search, using 12 NASA MDP datasets. The study also applies Random Forest classification and evaluates the model using AUC and t-Test. In addition, this study also compares the accuracy and precision values produced by each method. The results consistently show that the Copeland ranking method produces superior results compared to other ranking methods. The average AUC value obtained from the Copeland ranking method is 0.7496, higher than the Majority ranking method with an average AUC of 0.7416 and the Optimal Rank ranking method with an average AUC of 0.7343. These findings confirm that applying the Copeland ranking method in wrapper-based feature selection can enhance classification performance in software defect prediction using Random Forest compared to other ranking methods. The strength of the Copeland method lies in its ability to integrate rankings from various feature selection approaches and identify relevant features. The findings of this research demonstrate the potential of the Copeland ranking method as a reliable tool for ranking features obtained from various wrapper-based feature selection techniques. The implementation of this approach contributes to improved software defect prediction and provides new insights for the development of ranking methods in the future.

Keywords: Copeland Method, Feature Selection, Random Forest, Software Defect Prediction, Wrapper