Indonesia Agregasi Peringkat Berdasarkan Feature Filter Rangking Dalam Cross-Project Software Defects

Rudy Herteno, Mohammad Reza Faisal, Radityo Adi Nugroho, Friska Abadi, Setyo Wahyu SaputroNon-Scopus30 April 2025Afiliasi: ID

Abstrak

Software defects are a significant challenge in software engineering, as they can cause fatal damage if detected during system execution. This research focuses on Cross-Project Defect Prediction (CPDP), a methodology that utilizes historical data from different projects to improve defect prediction for the target project. However, CPDP is often constrained by data distribution mismatch and irrelevant high-dimensional features. To overcome this, we propose a novel approach with Feature Filter Ranking to reduce the dimensionality and overcome the imbalanced data, combined with Borda aggregation and classification algorithms KNN, Random Forest, Decision Tree, Logistic Regression, SVM, and Gardient Boosting. Experimental results show that the combination of 5 features on the NASA MDP dataset, 15 features on PROMISE, and 5 features on RELINK provides optimal performance. NASA MDP with KNN produces AUC 0.6600, PROMISE with KNN produces AUC 0.7000, and RELINK with KNN produces AUC 0.7167 for RELINK. From the average of all classification algorithms, it proves that KNN is more effective in improving the performance of software defect identification when viewed from the AUC. These results confirm that the integration of methods using CPDP with Feature Filter Ranking, Synthetic Data Vault, and Borda Aggregation helps to overcome the problem of data dimensionality and class imbalance, thus improving the process of predicting software defects.

Identifier

Scroll to Top