Model-based Software Defect Prediction from Software Quality Characterized Code Features by using Stacking Ensemble Learning

Abstract

Software defect prediction (SDP) is critical in guaranteeing software cost reduction and quality improvement while building a software system. Thus, software defect prediction in the early stages is an essential intrigue in the software engineering discipline. We proposed a stacking ensemble learning method to improve SDP performance based on software quality-defined code characteristics. Stacking combines the base classifiers by using a meta-classifier that learns base-classifiers output. It has certain advantages, such as easy implementation and combining classifiers by investigating various inducers. The proposed method's performance has been evaluated and compared with different Machine Learning (ML) classifiers on ivy2.0, tomcat, and velocity1.6 datasets available in PROMISE. The experimental findings revealed that the proposed approach has better prediction recall, accuracy, precision, AUC-ROC, and f-measure.