Predicting fault-prone software modules using feature selection and classification through data mining algorithms

Software defect detection has been an important topic of research in the field of software engineering for more than a decade. This research work aims to evaluate the performance of supervised machine learning techniques on predicting defective software through data mining algorithms. This paper places emphasis on the performance of classification algorithms in categorizing seven datasets (CM1, JM1, MW1, KC3, PC1, PC2, PC3 and PC4) under two classes namely Defective and Normal. In this study, publicly available data sets from different organizations are used. This permitted us to explore the impact of data from different sources on different processes for finding appropriate classification models. We propose a computational framework using data mining techniques to detect the existence of defects in software components. The framework comprises of data pre-processing, data classification and classifier evaluation. In this paper; we report the performance of twenty classification algorithms on seven publicly available datasets from the NASA MDP Repository. Random Tree Classification algorithm produced 100 percent accuracy in classifying the datasets and hence the features selected by this technique were considered to be the most significant features. The results were validated with suitable test data.