Effort-aware and just-in-time defect prediction with neural network

Area of Science:

• Software Engineering
• Machine Learning
• Software Quality Assurance

Background:

• Just-in-time (JIT) defect prediction aims to identify buggy code changes efficiently.
• Accurate defect prediction is crucial for optimizing inspection effort and improving software quality.
• Existing methods often struggle to balance defect likelihood with the cost of inspection.

Purpose of the Study:

• To propose a novel deep learning approach for effort-aware JIT defect prediction.
• To enhance the accuracy of predicting defects in source code changes.
• To improve the efficiency of software defect detection processes.

Main Methods:

• Preprocessing ten numerical code change metrics.
• Utilizing a neural network for feature selection and bug likelihood prediction.
• Calculating a benefit-cost ratio based on bug likelihood and code change size.
• Ranking code changes by their benefit-cost ratio.

Main Results:

• The proposed deep learning approach significantly outperforms state-of-the-art methods.
• Achieved an average improvement of 15.6% in recall.
• Demonstrated an 8.1% improvement in the 'popt' metric across subject projects.

Conclusions:

• Deep learning is effective for feature selection in effort-aware JIT defect prediction.
• The proposed method offers a superior approach to ranking code changes for defect inspection.
• This technique enhances the efficiency and effectiveness of software defect prediction.