Object Oriented Design Metrics for Software Defect Prediction: An Empirical Study

Abstract

Reliability of software in respect to defect prediction is a trending topic of study. It is observed that many effective categorical models rely on requirement and design phase metrics and few consistent models include metrics from design phase. It is also noted that most studies show qualitative advantages by using Early Software Defect Prediction models. It is necessary to validate software metrics in error prevision for object-driven methods by means of statistical methods and machine learning. The quality of the software product in a software organization is assured by the validation process. Object-oriented metrics play a key role in fault estimation. This paper discusses the use of Chidamber and Kemerer (CK) metrics, QMOOD, Size, Complexity and Martin's metrics for the assessment of software defects. In this study bug/defect is treated as a dependent variable and the considered metric fit as independent variables. For defect detection two data analysis techniques, logistic regression and Decision Tree, are applied, validated and their statistical efficacy and performance measures are discussed.