Analysis of Naive Bayes’ assumptions on software fault data: An empirical study

摘要

Software defect prediction is important for reducing test times by allocating testing resources effectively. In terms of predicting the defects in software, Naive Bayes outperforms a wide range of other methods. However, Naive Bayes assumes the ‘independence’ and ‘equal importance’ of attributes. In this work, we analyze these assumptions of Naive Bayes using public software defect data from NASA. Our analysis shows that independence assumption is not harmful for software defect data with PCA pre-processing. Our results also indicate that assigning weights to static code attributes may increase the prediction performance significantly, while removing the need for feature subset selection.