Development and evaluation of software fault prediction models using object oriented and code smell metrics

Abstract

Software quality is an essential feature of any software; therefore, organizations are directing a lot of resources in developing good quality software. The growing complexity of today s software has led to an increase in the number of lines of code, which has increased the number of modules. If testing is not done thoroughly then some modules are likely to be prone to issues. In the real world, it is impossible to test every module of any software in a white box environment so, software testing has to be performed on a case-by-case basis for the components that are prone to failure by identifying the likely faultprone modules of the software. Early detection and correction of flaws result in higher quality and lower costs of software. As a result, software fault prediction approaches have gotten a lot of attention in the software engineering field. The major contribution of the thesis pertains to investigate variety of machine learning techniques for software fault prediction using object-oriented and code smell metrics. Statistical tests have been used to report the empirical results of the current work. The study extensively analysed five data sets (Dr Java, PMD, EMMA, FindBugs and Trove) that were gathered manually and were then mined using a metrics extraction tool. The current work has been primarily divided into four sections. In the first part, categories of classifiers, namely Tree, KNN, Regression, SVM and Ensemble, were compared using state-of-the-art measurements for software fault prediction. The findings of the current work concluded that Ensemble and SVM are the best categories of classification. In the second section, the dimensional reduction was accomplished without compromising the performance of the classifiers by employing diverse Feature Selection Techniques. The study also compared the performance of these classifiers with and without Feature Selection Techniques. The work further suggested enhanced Feature Selection by employing three aggregators, namely Union, Intersection...