Exploring the impact of data preprocessing techniques on composite classifier algorithms in cross-project defect prediction

Abstract

Success in software projects is now an important challenge. The main focus of the engineering community is to predict software defects based on the history of classes and other code elements. However, these software defect prediction techniques are effective only as long as there is enough data to train the prediction model. To mitigate this problem, cross-project defect prediction is used. The purpose of this research investigation is twofold: first, to replicate the experiments in the original paper proposal, and second, to investigate other settings regarding defect prediction with the aim of providing new insights and results regarding the best approach. In this study, three composite algorithms, namely AvgVoting, MaxVoting and Bagging are used. These algorithms integrate multiple machine classifiers to improve cross-project defect prediction. The experiments use pre-processed methods (normalization and standardization) and also feature selection. The results of the replicated experiments confirm the original findings when using raw data for all three methods. When normalization is applied, better results than in the original paper are obtained. Even better results are obtained when feature selection is used. In the original paper, the MaxVoting approach shows the best performance in terms of the F-measure, and BaggingJ48 shows the best performance in terms of cost-effectiveness. The same results in terms of F-measure were obtained in the current experiments: best MaxVoting, followed by AvgVoting and then by BaggingJ48. Our results emphasize the previously obtained outcome; the original study is confirmed when using raw data. Moreover, we obtained better results when using preprocessing and feature selection.