EUVIMEDEUVIMED entdecken
Crossref

A Machine Learning-Based Framework for Software Defect Prediction Using Structural Software Metrics: An Empirical Analysis

Isaiah Ifeanyi Nweze, Paul Maduabuchi Agu, Ezekiel Nwibo Gabriel, Charles Ugwute, Chukwuka Abraham Nwovu, Boniface Mbalaba Ofoke

International Journal of Innovative Science and Research Technology · 26.05.2026 · 1642

DOI: 10.38124/ijisrt/26may670 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Innovative Science and Research Technology, online erschienen am 26.05.2026. Verfasst von Isaiah Ifeanyi Nweze und 5 weiteren Autor:innen. Die Arbeit nennt 27 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Software defects are a critical challenge in modern software engineering systems due to the growing complexity of the structure and interdependence of objects in object-oriented system designs, which negatively affect the maintenance costs and reliability of the system. This paper introduces a machine learning-based system that predicts software defects using structural software metrics. A structured dataset of 145 software modules of 94 numerical features was analysed using six major metrics: Coupling Between Objects (CBO), Depth of Inheritance Tree (DIT), Lack of Cohesion of Methods (LCOM), Number of Children (NOC), Response for Class (RFC) and Weighted Methods per Class (WMC). The statistical analysis indicated substantial variation and skewness in complexity-related measures, particularly CBO and RFC, suggesting the presence of structural outliers. Correlation analysis revealed a significant association among coupling, cohesion, and response measures, indicating that defect proneness is not affected by single factors but rather by interacting structural properties. The dataset had a moderate class imbalance, with most modules being non-defective. To compare the predictive performance, three supervised machine learning models (Decision Tree, Random Forest and Logistic Regression) were trained using a stratified 70:30 train-test split. The Random Forest model achieved the highest overall performance with an accuracy of 72.73%, precision of 75%, recall of 50%, and an F1-score of 0.60, reflecting a good balance between classification accuracy and generalisation. Logistic Regression was more precise but had less recall, whereas the Decision Tree model was less accurate but more interpretable. In general, the results show that statistical analysis and machine learning provide an efficient approach to early defect detection. The paper highlights the importance of structural complexity measures in identifying defect-sensitive modules. It advocates using an ensemble learning algorithm to enhance software quality assurance and overall system reliability.

Zitieren

APA

Nweze, I. I., Agu, P. M., Gabriel, E. N., Ugwute, C., Nwovu, C. A., & Ofoke, B. M. (2026). A Machine Learning-Based Framework for Software Defect Prediction Using Structural Software Metrics: An Empirical Analysis. International Journal of Innovative Science and Research Technology, 1642. https://doi.org/10.38124/ijisrt/26may670

Vancouver

Nweze II, Agu PM, Gabriel EN, Ugwute C, Nwovu CA, Ofoke BM. A Machine Learning-Based Framework for Software Defect Prediction Using Structural Software Metrics: An Empirical Analysis. International Journal of Innovative Science and Research Technology. 2026:1642. doi:10.38124/ijisrt/26may670

BibTeX
@article{Nweze2026,
  author = {Nweze, Isaiah Ifeanyi and Agu, Paul Maduabuchi and Gabriel, Ezekiel Nwibo and Ugwute, Charles and Nwovu, Chukwuka Abraham and Ofoke, Boniface Mbalaba},
  title = {A Machine Learning-Based Framework for Software Defect Prediction Using Structural Software Metrics: An Empirical Analysis},
  journal = {International Journal of Innovative Science and Research Technology},
  year = {2026},
  pages = {1642},
  doi = {10.38124/ijisrt/26may670}
}
Kostenlos starten →