Detecting concept drift in just-in-time software defect prediction using model interpretation

Key Points

• Concept drift negatively affects the accuracy of software defect prediction models, complicating their effectiveness.
• The study identifies key parameters that influence model performance over time, underscoring the need for timely adjustments.
• Using advanced model interpretation techniques, the research detects shifts in data distribution relevant to defect prediction.
• Highlighting the importance of adaptive algorithms, the findings suggest improved methods could enhance prediction reliability in real-world applications.