View in EDS

Interpretable Software Defect Prediction Using Hierarchical Multi-head Attention Networks on Defect Log Data.

Saved in:
Bibliographic Details
Title: Interpretable Software Defect Prediction Using Hierarchical Multi-head Attention Networks on Defect Log Data.
Authors: Gottumukkala, Devi Priya, P. V. G. D., Prasad Reddy, Rao, S. Krishna
Source: International Journal of Intelligent Engineering & Systems; 2026, Vol. 19 Issue 1, p128-145, 18p
Subject Terms: DEEP learning, DEFECT tracking (Computer software development), INFORMATION processing, FORECASTING, FAILURE analysis
Abstract: Software defect prediction (SDP) helps identify modules with a high probability of defects prior in the development lifecycle to improve software quality and curb repair expenditure. Conventional approaches of defect prediction rely on source code metrics and handcrafted features that are not effective in representing defect log semantics. Defect logs contain rich information, and this work addresses the semantic gap by proposing a deep learning model H-MHAN that employs multi-head attention mechanisms to model defect data at the word and sentence levels. The suggested model improves the prediction of a defect-prone module by focusing the model on the most descriptive textual signals using hierarchical attention, a method that is also helpful in explaining the model's decisions. Real-world experiments on an industrial dataset demonstrate strong performance, achieving 0.9886 accuracy, 0.9922 F1-score, and 0.9971 ROC AUC in single hold-out evaluation. To prevent overfitting and promote robustness, repeated k-fold cross-validation was run, and reported mean values with standard deviation. Statistical significance testing established improvements over the TF--IDF + Logistic Regression baseline. Hierarchical attention promotes interpretability by emphasizing effective segments of defect logs, thus enhancing developer confidence and facilitating defect triaging. The above findings establish the viability of H-MHAN for real-world deployment in fault prediction tasks. [ABSTRACT FROM AUTHOR]
Copyright of International Journal of Intelligent Engineering & Systems is the property of Intelligent Networks & Systems Society and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database: Complementary Index
Description
Abstract:Software defect prediction (SDP) helps identify modules with a high probability of defects prior in the development lifecycle to improve software quality and curb repair expenditure. Conventional approaches of defect prediction rely on source code metrics and handcrafted features that are not effective in representing defect log semantics. Defect logs contain rich information, and this work addresses the semantic gap by proposing a deep learning model H-MHAN that employs multi-head attention mechanisms to model defect data at the word and sentence levels. The suggested model improves the prediction of a defect-prone module by focusing the model on the most descriptive textual signals using hierarchical attention, a method that is also helpful in explaining the model's decisions. Real-world experiments on an industrial dataset demonstrate strong performance, achieving 0.9886 accuracy, 0.9922 F1-score, and 0.9971 ROC AUC in single hold-out evaluation. To prevent overfitting and promote robustness, repeated k-fold cross-validation was run, and reported mean values with standard deviation. Statistical significance testing established improvements over the TF--IDF + Logistic Regression baseline. Hierarchical attention promotes interpretability by emphasizing effective segments of defect logs, thus enhancing developer confidence and facilitating defect triaging. The above findings establish the viability of H-MHAN for real-world deployment in fault prediction tasks. [ABSTRACT FROM AUTHOR]
ISSN:2185310X
DOI:10.22266/ijies2026.0131.09