Software Defect Prediction Based on Ensemble Learning

Software defect prediction is one of the important ways to guarantee the quality of software systems. Combining various algorithms in machine learning to predict software defects has become a hot topic in the current study. The paper uses the datasets of MDP as the experimental research objects and takes ensemble learning as research focus to construct software defect prediction model. With experimenting five different types of ensemble algorithms and analyzing the features and procedures, this paper discusses the best ensemble algorithm which is Random Forest through experimental comparison. Then we utilize the SMOTE over-sampling and Resample methods to improve the quality of datasets to build a complete new software defect prediction model. Therefore, the results show that the model can improve defect classification performance effectively.