Design of a Hybrid Machine Learning Base-Classifiers for Software Defect Prediction

Abstract

Machine learning (ML) classifiers have attracted the research community's attention in the field of software defect prediction (SDP) in the last decade. Several comparative studies confirmed that algorithms like Naí¯ve Bayes, Decision tree, or Random Forest have been used with good performance. Recently, several hybrid classifiers were also introduced in SDP. However, many of those can only provide a category for a given new sample. Instead, this paper proposes a methodology to build a hybrid of four (4) ML Base-classifiers made up of Gaussian Naive Bayes, Bernoulli Naive Bayes, Random Forest, and support vector machine (SVM). The study carefully ensemble the selected machine learning models with efficient feature selection to address these issues and mitigate their effects on the defect prediction performance using dataset CM1 from PROMISE repository. The study outcome is expected to show promising SDP performance in terms of F1-score compared to the benchmark work.