EMPIRICAL VALIDATION OF OBJECT ORIENTED METRICS USING MACHINE LEARNING METHODS

Abstract

Complex and advanced software systems are more prone to faults and result in greater maintenance cost in later stages of software development cycle. With the help of this study we suggest the importance of machine learning algorithms in detection of fault proneness in software systems results in early stages of software development life cycle. We concentrated on the use of machine learning methods and used them for empirically validating object-oriented design metrics, Chidamber et al. [1], for the purpose of predicting fault proneness. We have used open source project developed in Java language, “MX4J” and “Synapse 1.2”, as the base of our empirical study. The defect prediction models developed using machine learning methods are used to compute and evaluate performance of these models. We evaluated the performance using Receiver Operating Characteristic (ROC) analysis. We used tools such as Weka and SPSS for the purpose of generating data distribution and ROC curve. As per the ROC analysis for both the projects, machine learning methods LogitBoost and Bagging show better performance as compared to other machine learning methods.