Practitioners' Perceptions of the Goals and Visual Explanations of Defect Prediction Models

Practitioners find defect prediction models valuable: 82%-84% of surveyed software developers and managers perceived the three main goals of defect prediction models as useful, suggesting these AI tools address real needs in software quality assurance and resource planning.

Visual explanations matter for trust and adoption: The study found that practitioners have distinct preferences for different explanation techniques (LIME for understanding individual predictions, ANOVA/VarImp for understanding general patterns), indicating that how AI models are explained significantly influences whether professionals will actually use them in practice.

Explainable AI techniques can bridge the gap between developers and AI: The research demonstrates that model-agnostic explanation techniques from the explainable AI field can help software practitioners understand why defect prediction models make specific recommendations, addressing a key barrier to responsible AI deployment in real-world software development.

One-size-fits-all explanations don't work: Different practitioners need different types of explanations depending on their goals—whether they're trying to understand a specific prediction or identify general patterns—suggesting that responsible AI implementation requires tailoring explanations to user needs rather than applying uniform approaches.