An Ensemble DeepBoost Classifier for Software Defect Prediction

Abstract

The main objective of a software development team is to have maximum customer defects in the software will reduce its quality. Thereby increasing its development cost. Several algorithms have been proposed for predicting software defects. But most of these algorithms are not appropriate when the dataset is imbalanced. In this paper an Ensemble DeepBoost Classifier (EDC) is built to predict the software defects effectively by addressing two major issues – curse of dimensionality and class distribution imbalance problem. Firstly, EDC uses Genetic Algorithm (GA) to find out the features that are relevant for software defect prediction. Thus, achieving dimensionality reduction. Later it uses Safe Line SMOTE (SLS) algorithm to achieve equal class distribution. Finally, it uses DeepBoost algorithm to predict whether the samples are defective or not based on the historical software defect data. The experiment was carried out on 7 PROMISE repository datasets and the results of EDC were compared with similar algorithms. The experimental results indicate that EDC has outperformed various existing algorithms in most evaluation metrics.