A Hybrid Statistical-Semantic Framework with Domain Adaptation for Cross-Project Software Fault Prediction
Computer and Decision Making: An International Journal · 08.11.2026 · 3 (7S) · 21-49
DOI: 10.1922/comdem.v3i7s.19073 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 08.11.2026. Verfasst von Navneet Malik und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Software fault prediction is essential for improving the quality, reliability, and maintainability of complex object-oriented software systems. Existing Machine Learning (ML) approaches have achieved promising results. However, limitations remain in integrating statistical and semantic software features, handling crosses-project distribution differences, and providing interpretable predictions. Therefore, this study aims to develop an intelligent ML framework for effective Cross-Project Software Fault Prediction (CPSFP). The framework was evaluated using the PROMISE Repository dataset comprising six software projects. The proposed framework integrates Chidamber and Kemerer (CK), Halstead, and McCabe software metrics with Abstract Syntax Tree (AST), Control Flow Graph (CFG), and Program Dependency Graph (PDG) representations. GraphCodeBERT and Heterogeneous Graph Transformer (HGT) are employed for semantic feature learning, while Cross-Attention Fusion combines statistical and semantic representations. SHapley Additive exPlanations (SHAP)-Guided Recursive Feature Elimination (RFE) is applied for feature selection and explainability, followed by Domain Adversarial Neural Network (DANN) for domain-adaptive fault classification. The framework was evaluated using cross-project validation and ablation analysis to assess its predictive performance and component contributions. The ablation study showed progressive improvements with the integration of the proposed components. The current experimental evaluation achieved 99.83% accuracy and 99.83% F1-score, indicating strong fault classification performance. Overall, the proposed framework effectively combines statistical, semantic, feature-fusion, explainability, and domain-adaptation techniques for CPSFP, with potential to support reliable and interpretable software quality assessment and informed decision-making across diverse object-oriented software projects.
Zitieren
APA
Malik, N., Mishra, D. A. K., & Verma, D. D. K. (2026). A Hybrid Statistical-Semantic Framework with Domain Adaptation for Cross-Project Software Fault Prediction. Computer and Decision Making: An International Journal, 3(7S), 21-49. https://doi.org/10.1922/comdem.v3i7s.19073
Vancouver
Malik N, Mishra DAK, Verma DDK. A Hybrid Statistical-Semantic Framework with Domain Adaptation for Cross-Project Software Fault Prediction. Computer and Decision Making: An International Journal. 2026;3(7S):21-49. doi:10.1922/comdem.v3i7s.19073
BibTeX
@article{Malik2026,
author = {Malik, Navneet and Mishra, Dr Anil Kumar and Verma, Dr. Deepak Kumar},
title = {A Hybrid Statistical-Semantic Framework with Domain Adaptation for Cross-Project Software Fault Prediction},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {7S},
pages = {21--49},
doi = {10.1922/comdem.v3i7s.19073}
}
EUVIMED