Assessment of defect prediction models using machine learning techniques for object-oriented systems

Abstract

Software development is an essential field today. The advancement in software systems leads to risk of them being exposed to defects. It is important to predict the defects well in advance in order to help the researchers and developers to build cost effective and reliable software. Defect prediction models extract information about the software from its past releases and predict the occurrence of defects in future releases. A number of Machine Learning (ML) algorithms proposed and used in the literature to efficiently develop defect prediction models. What is required is the comparison of these ML techniques to quantify the advantage in performance of using a particular technique over another. This study scrutinizes and compares the performances of 17 ML techniques on the selected datasets to find the ML technique which gives the best performance for determining defect prone classes in an Object-Oriented(OO) software. Also, the superiority of the best ML technique is statistically evaluated. The result of this study demonstrates the predictive capability of ML techniques and advocates the use of Bagging as the best ML technique for defect prediction.