Analysis and Selection of Optimized Machine Learning Techniques for Software Bug Prediction

Abstract

Software Bug Prediction is an active research area and is being widely explored with the help of Machine Learning. Since bug prediction is now considered as an important measure of SDLC, we need to have optimized techniques for making predictive models. Presently transfer learning and ensemble learning approaches are being researched much. However, previous studies are not sufficient in this regard. So in this paper a framework is created by using multiple techniques to explore their effectiveness when combined in one model. The techniques involved feature selection which is used to reduce the dimensionality and redundancy of features and select only the relevant ones; transfer learning is used to train and test the model on different datasets to analyze how much of the learning is passed to other dataset; and ensemble method is utilized to explore the increase in performance upon combining multiple classifiers in a model. Four NASA and four Promise datasets are used in the study, the results of which show an increase in the performance of the model by providing better AUC-ROC values when different classifiers were combined in the model. Thus revealing that use of amalgam of techniques such as used in this study, feature selection, transfer learning and ensemble methods prove helpful in optimizing the software bug prediction models and provide high performing, useful end model.