Scholarlink

<p>Software defect prediction plays a critical role in improving software quality by enabling early identification of fault-prone modules. However, existing prediction approaches often suffer from limited generalization under class imbalance, lack of interpretability, and insufficient practical usability for real-world development environments. To address these challenges, this study proposes a graphical user interface (GUI)–integrated explainable multi-branch multi-fusion GoogleNet framework, referred to as Xai-MBMFG, for software defect prediction. The proposed framework combines multiple complementary learning branches, including dilated convolutions, self-attention mechanisms, and enhanced convolutional layers, to capture diverse defect-related patterns. In addition, statistical feature measures such as entropy, cosine coefficient, and Bhattacharyya coefficient are employed to reduce uncertainty and improve feature discrimination prior to deep learning. Model transparency is enhanced through the integration of explainable AI techniques, specifically LIME and SHAP, which provide localized explanations for individual predictions. The effectiveness of the proposed approach is evaluated on three widely used benchmark datasets—AEEM, NASA, and PROMISE—using multiple performance metrics. The implementation is done by the PYTHON. The experimental results demonstrate that the segmentation model achieves the highest precision (0.9857). Ablation and cross-validation studies further confirm the contribution and stability of each architectural component.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An explainable AI-based interpreter for deep-dual-wave based Unet framework with a cross-attention layer-based dental caries segmentation and classification

  • 期刊论文

  • 发表日期：

    2026年3月7日

查看全文
  • Vivekanand Aelgani,
  • Akansha Singh,
  • V. A. Narayana

摘要

Software defect prediction plays a critical role in improving software quality by enabling early identification of fault-prone modules. However, existing prediction approaches often suffer from limited generalization under class imbalance, lack of interpretability, and insufficient practical usability for real-world development environments. To address these challenges, this study proposes a graphical user interface (GUI)–integrated explainable multi-branch multi-fusion GoogleNet framework, referred to as Xai-MBMFG, for software defect prediction. The proposed framework combines multiple complementary learning branches, including dilated convolutions, self-attention mechanisms, and enhanced convolutional layers, to capture diverse defect-related patterns. In addition, statistical feature measures such as entropy, cosine coefficient, and Bhattacharyya coefficient are employed to reduce uncertainty and improve feature discrimination prior to deep learning. Model transparency is enhanced through the integration of explainable AI techniques, specifically LIME and SHAP, which provide localized explanations for individual predictions. The effectiveness of the proposed approach is evaluated on three widely used benchmark datasets—AEEM, NASA, and PROMISE—using multiple performance metrics. The implementation is done by the PYTHON. The experimental results demonstrate that the segmentation model achieves the highest precision (0.9857). Ablation and cross-validation studies further confirm the contribution and stability of each architectural component.