Faulty Classes Prediction in Object-Oriented Programming Using Composed Dagging Technique

Abstract

Class is one of the fundamental concepts of the object-oriented paradigm and has been scrutinized since the developers moved on from procedural programming design. In software fault prediction, the legalization of software metrics is essential. As a handful of software metrics suites exist, it is a very hard task to predict the defective classes flawlessly using a particular set of metrics suites. However, it is a rational approach to use only the object-oriented metrics that are directly relatable to the class definitions in the code that helps the developers foresee the errors in defining the classes and minimize the errors as much as possible. This paper utilized twelve object-oriented metrics selected from various metrics suites. The dagging ensemble model is merged with three well-known classification algorithms (Naive Bayes, Multilayer Perceptron, J48 Decision Tree) individually and applied to twelve java projects. The study depicts that the proposed ensemble method gives improved outcomes that are statistically significant when merged with Naive Bayes and Multilayer Perceptron. The proposed ensemble method shows improvements up to 12.5% in accuracy and 15% in F-Score.