Comparative analysis of software fault prediction using various categories of classifiers

Inderpreet Kaur, Arvinder Kaur

Abstract: The quality of the software being developed varies with the size and complexity of the software. It is a matter of concern in software development as it impairs the faith of customers on the software companies. The quality of software can be improved if the prediction of faults and flaws in it are done in the early phases of the software development and thus reducing the resources to be used in the testing phase. The rise in the use of Object-Oriented technology for developing software has paved the way for considering the Object-Oriented metrics for software fault prediction. Numerous machine learning and statistical techniques have been used to predict the defects in software using these software metrics as independent variables and bug proneness as dependent variable. Our work aims at finding the best category and hence the best classifier for classification of faults. This work uses twenty-one classifiers belonging to five categories of classification on five open source software having Object-Oriented metrics. The classification LearnerApp of MATLAB has been used to evaluate various classification models. The work proposes the use of Ensemble and SVM techniques over KNN, Regression, and Tree. The bagged trees (ensemble) and cubic (SVM) are found to be the best predictors amongst the twenty-one classifiers.