Layered Defect Prediction Model for Software Product Lines Using Feature Fusion and Ensemble Classification
محورهای موضوعی : Journal of Computer & Robotics
Mehdi Habibzadeh-khameneh
1
,
Akbar Nabiollahi Najafabadi
2
,
Reza Tavoli
3
,
Hamid Rastegari
4
1 - Faculty of Computer Engineering, Najafabad branch, Islamic Azad University, Najafabad, Iran
2 - Faculty of Computer Engineering, Najafabad branch, Islamic Azad University, Najafabad, Iran
3 - Faculty ofDepartment of Mathematics, Islamic Azad University, Chalous Branch (IAUC)
4 - Faculty of Computer Engineering, Najafabad branch, Islamic Azad University, Najafabad, Iran
کلید واژه: Software Product lines, Defect Prediction, Metaheuristic Algorithms, Ensemble Learning, Feature Fusion,
چکیده مقاله :
Software defect prediction plays a critical role in ensuring the quality and reliability of modern software systems, particularly within Software Product Lines (SPLs), where numerous feature combinations and collaborative development efforts often result in flawed feature selections and increased risk of defects. This study presents a novel three-layered architecture designed to enhance defect prediction accuracy. The architecture consists of: (1) dataset preprocessing and optimal feature extraction using metaheuristic algorithms, (2) feature fusion to enrich the input representation, and (3) final classification through a stacking-based ensemble learning strategy. The proposed method was rigorously evaluated on benchmark datasets from NASA (CM1, JM1, KC1, PC1) and the Linux Variable Analysis Tools (LVAT) repository (LTS1, LTM2, LTL3, LTV4). It achieved accuracy rates of 98.62%, 97.81%, 98.59%, and 98.71% on the NASA datasets, and 95.1%, 94.2%, 97.3%, and 99.4% on the LVAT datasets, respectively. These results demonstrate that the proposed approach consistently outperforms existing methods across diverse SPL scenarios.
Software defect prediction plays a critical role in ensuring the quality and reliability of modern software systems, particularly within Software Product Lines (SPLs), where numerous feature combinations and collaborative development efforts often result in flawed feature selections and increased risk of defects. This study presents a novel three-layered architecture designed to enhance defect prediction accuracy. The architecture consists of: (1) dataset preprocessing and optimal feature extraction using metaheuristic algorithms, (2) feature fusion to enrich the input representation, and (3) final classification through a stacking-based ensemble learning strategy. The proposed method was rigorously evaluated on benchmark datasets from NASA (CM1, JM1, KC1, PC1) and the Linux Variable Analysis Tools (LVAT) repository (LTS1, LTM2, LTL3, LTV4). It achieved accuracy rates of 98.62%, 97.81%, 98.59%, and 98.71% on the NASA datasets, and 95.1%, 94.2%, 97.3%, and 99.4% on the LVAT datasets, respectively. These results demonstrate that the proposed approach consistently outperforms existing methods across diverse SPL scenarios.
