Improvement in Software Defect Prediction Outcome Using Principal Component Analysis and Ensemble Machine Learning Algorithms

Abstract

Improving customer experience is the focus of IT Industry. It is no longer about customer satisfaction, but it is about creating memorable experiences which will help build loyal customers. Hence it is extremely critical to release defect free software. While machine learning techniques were widely used for prediction modelling, creating a reliable predictor which can perform satisfactorily is always a challenge. In this paper, we have proposed a framework using PCA for feature selection and ensemble machine learning algorithms with stratified 10-fold cross validation for building the classification model. The proposed model is tested using 5 projects from NASA Metrics Data program and 4 ensemble machine learning algorithms. Our results show that the prediction accuracy is improved by 0.6% when the reduced dataset is used for classification than using the whole dataset. In comparison with previous research studies, our framework has shown an average of 4.2% increase in performance.