Superposed Naive Bayes for Accurate and Interpretable Prediction

Abstract

Background: Data mining and machine learning techniques have been widely applied in software engineering research. However, past research has mainly focused on only prediction accuracy. Aim: The interpretability of prediction results should be accorded greater emphasis in software engineering research. A prediction model that has high accuracy and explanatory power is required. Method: We propose a new algorithm of naïve Bayes ensemble, called superposed naïve Bayes (SNB), which firstly builds an ensemble model with high prediction accuracy and then transforms it into an interpretable naïve Bayes model. Results: We conducted an experiment with the NASA MDP datasets, in which the performance and interpretability of the proposed method were compared with those of other classification techniques. The results of the experiment indicate that the proposed method can produce balanced outputs that satisfy both performance and interpretability criteria. Conclusion: We confirmed the effectiveness of the proposed method in an experiment using software defect data. The model can be extensively applied to other application areas, where both performance and interpretability are required.