Voting Based Ensemble Classification for Software Defect Prediction

Abstract

Fault Prediction procedures are meant to help focus on software testing and troubleshooting; they can caution developers on programming segments that appear to be defective. Here, a Voting Based Ensemble Classification is proposed in which we apply feature selection on a preprocessed data utilizing two approaches, which include a Wrapper-based and Heuristic-based approach using python and WEKA respectively. After which, we train a classifier built on the selected attributes using the Voting Based Ensemble Learning Algorithm where predictions from multiple models are combined. We use three base learners which include the Adaboost classifier, Random Forest classifier, and Naive Bayes classifier. This paper embraces NASA datasets to check the exhibition of this design. The paper demonstrates that preprocessing technique using the Wrapper-based approach outperforms preprocessing done using the Heuristic-based approach. It also proved that the new Voting Based Ensemble Learning Algorithm consisting of Random Forests (RF), Adaboost, and Naive Bayes is a better approach than other algorithms for Software Defect Prediction that currently exist.