Open access Jurnal SINTA 2
Multi-Criteria Decision Making dalam Seleksi Fitur Ensemble untuk Prediksi Cacat Perangkat Lunak
Jurnal Teknologi Informasi dan Ilmu Komputer (JTIIK) · Vol. 12 No. 6 (2025) · pp. 1385–1394
- ISSN
- 2528-6579, 2355-7699
- Published
- 2025-12-17
- Publisher
- Fakultas Ilmu Komputer, Universitas Brawijaya
- Language
- id
Abstract
Bahasa Indonesia
Prediksi cacat perangkat lunak merupakan upaya strategis dalam meningkatkan kualitas produk melalui identifikasi dini modul yang berpotensi cacat. Kinerja prediksi dipengaruhi oleh pemilihan fitur, karena informasi yang berlebihan dan tidak relevan dapat mempengaruhi kualitas pembelajaran model. Seleksi fitur ensemble dinilai efektif dalam menyeleksi fitur yang relevan dengan menggabungkan beberapa metode seleksi fitur berbasis filter. Diperlukan mekanisme integrasi untuk menyatukan hasil dari empat teknik filter—Mutual Information, Fisher Score, Uncertainty dan Relief. Penelitian ini membandingkan empat metode Multi‑Criteria Decision Making—TOPSIS, VIKOR, EDAS, dan WASPAS—yang bekerja dengan merangking nilai relevansi fitur hasil seleksi filter tersebut. Sepuluh fitur teratas dari tiap metode kemudian dievaluasi menggunakan model Random Forest dengan metrik AUC melalui K‑Fold cross‑validation. Dari 12 dataset NASA MDP yang diuji, TOPSIS menunjukkan kinerja paling konsisten dan terbaik dengan nilai rata-rata AUC sebesar 0,8038. Temuan ini menegaskan pentingnya pemilihan metode integrasi yang tepat dalam meningkatkan akurasi prediksi cacat perangkat lunak dan memberikan panduan bagi pengembangan model yang lebih efektif. Abstract Software defect prediction is a strategic effort to improve product quality through early identification of potentially defective modules. Prediction performance is influenced by feature selection, because redundant and irrelevant information can affect the quality of model learning. Ensemble feature selection is considered effective in selecting relevant features by combining several filter-based feature selection methods. An integration mechanism is needed to unify the results of four filter techniques—Mutual Information, Fisher Score, Uncertainty and Relief. This study compares four Multi-Criteria Decision Making methods—TOPSIS, VIKOR, EDAS, and WASPAS—which work by ranking the relevance values of the filter-selected features. The top ten features from each method are then evaluated using the Random Forest model with the AUC metric through K-Fold cross-validation. Of the 12 NASA MDP datasets tested, TOPSIS showed the most consistent and best performance with an average AUC value of 0.8038. These findings emphasize the importance of choosing the right integration method in improving the accuracy of software defect prediction and provide guidance for the development of more effective models.
English
Prediksi cacat perangkat lunak merupakan upaya strategis dalam meningkatkan kualitas produk melalui identifikasi dini modul yang berpotensi cacat. Kinerja prediksi dipengaruhi oleh pemilihan fitur, karena informasi yang berlebihan dan tidak relevan dapat mempengaruhi kualitas pembelajaran model. Seleksi fitur ensemble dinilai efektif dalam menyeleksi fitur yang relevan dengan menggabungkan beberapa metode seleksi fitur berbasis filter. Diperlukan mekanisme integrasi untuk menyatukan hasil dari empat teknik filter—Mutual Information, Fisher Score, Uncertainty dan Relief. Penelitian ini membandingkan empat metode Multi‑Criteria Decision Making—TOPSIS, VIKOR, EDAS, dan WASPAS—yang bekerja dengan merangking nilai relevansi fitur hasil seleksi filter tersebut. Sepuluh fitur teratas dari tiap metode kemudian dievaluasi menggunakan model Random Forest dengan metrik AUC melalui K‑Fold cross‑validation. Dari 12 dataset NASA MDP yang diuji, TOPSIS menunjukkan kinerja paling konsisten dan terbaik dengan nilai rata-rata AUC sebesar 0,8038. Temuan ini menegaskan pentingnya pemilihan metode integrasi yang tepat dalam meningkatkan akurasi prediksi cacat perangkat lunak dan memberikan panduan bagi pengembangan model yang lebih efektif. Abstract Software defect prediction is a strategic effort to improve product quality through early identification of potentially defective modules. Prediction performance is influenced by feature selection, because redundant and irrelevant information can affect the quality of model learning. Ensemble feature selection is considered effective in selecting relevant features by combining several filter-based feature selection methods. An integration mechanism is needed to unify the results of four filter techniques—Mutual Information, Fisher Score, Uncertainty and Relief. This study compares four Multi-Criteria Decision Making methods—TOPSIS, VIKOR, EDAS, and WASPAS—which work by ranking the relevance values of the filter-selected features. The top ten features from each method are then evaluated using the Random Forest model with the AUC metric through K-Fold cross-validation. Of the 12 NASA MDP datasets tested, TOPSIS showed the most consistent and best performance with an average AUC value of 0.8038. These findings emphasize the importance of choosing the right integration method in improving the accuracy of software defect prediction and provide guidance for the development of more effective models.
Keywords
Collaboration network of the authors
27 authors · 87 co-authorship links · 4 across different organisations
Hover or tap a name to highlight its co-authors. Double-click to open the author's page.
- Lambung Mangkurat University
- Universitas Mercu Buana Yogyakarta
- Diponegoro University
- Affiliation unknown
- Different organisations
- Same organisation
- Affiliation unknown
- Thicker line = more joint articles · bigger circle = more articles
Show as table
| Co-author | Organisation | Joint articles |
|---|---|---|
| Dwi Kartini | Lambung Mangkurat University | 5 |
| Muhammad Itqan Mazdadi | Lambung Mangkurat University | 4 |
| Radityo Adi Nugroho | Lambung Mangkurat University | 4 |
| Setyo Wahyu Saputro | Lambung Mangkurat University | 3 |
| Andi Farmadi | Lambung Mangkurat University | 2 |
| Mohammad Reza Faisal | Lambung Mangkurat University | 2 |
| Irwan Budiman | Lambung Mangkurat University | 2 |
| Putri Nabella | Affiliation unknown | 2 |
| Putri Nabella | Lambung Mangkurat University | 2 |
| Adela Putri Ariyanti | Lambung Mangkurat University | 1 |
| Muliadi Muliadi | Lambung Mangkurat University | 1 |
| Salsha Farahdiba | Lambung Mangkurat University | 1 |
| Triando Hamonangan Saragih | Lambung Mangkurat University | 1 |
| Muhammad Hevny Rizky | Lambung Mangkurat University | 1 |
| Nitami Lestari Putri | Diponegoro University | 1 |
| Muhammad Rifqi | Universitas Mercu Buana Yogyakarta | 1 |
| Deni Kurnia | Lambung Mangkurat University | 1 |
| Siti Roziana Azizah | Lambung Mangkurat University | 1 |
| Syarif Maulana | Affiliation unknown | 1 |
| Dodon Turianto Nugrahadi | Affiliation unknown | 1 |
| Mohammad Reza Faisal | Affiliation unknown | 1 |
| Reina Alya Rahma | Lambung Mangkurat University | 1 |
Affiliations come from OpenAlex; journal OAI-PMH metadata carries none. Authors without an OpenAlex ID are matched by name, which can merge or split people.