A trustworthy hybrid model for transparent software defect prediction: SPAM-XAI.

Area of Science:

• Software Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Software development complexity is increasing, leading to challenges in maintaining quality.
• Software defect prediction (SDP) is crucial for early detection of faulty modules and cost reduction.
• Existing SDP models face issues with imbalanced data, high dimensionality, overfitting, and lack of transparency.

Purpose of the Study:

• To address the limitations of traditional SDP models.
• To propose a hybrid model, SPAM-XAI, integrating sampling, feature selection, and eXplainable-AI (XAI).
• To enhance the robustness, transparency, and trustworthiness of SDP in the Software Development Life Cycle (SDLC).

Main Methods:

• Developed SPAM-XAI, a hybrid model combining novel sampling and feature selection techniques.
• Integrated eXplainable-AI (XAI) algorithms for model transparency and interpretability.
• Reduced feature dimensionality and optimized model complexity to improve efficiency.

Main Results:

• SPAM-XAI demonstrated superior performance on NASA PROMISE datasets.
• Achieved high accuracy: 98.13% (CM1), 96.00% (PC1), and 98.65% (PC2).
• Outperformed state-of-the-art and baseline models across multiple evaluation metrics.

Conclusions:

• SPAM-XAI effectively addresses challenges in SDP, including data imbalance and model interpretability.
• The model enhances transparency, facilitating understanding of feature-error interactions.
• SPAM-XAI improves decision-making and trustworthiness within the SDLC.