Software Defect Prediction Model Based on Improved BP Neural Network

This paper proposes a software defect prediction algorithm based on improved BP neural network, which can effectively improve the prediction accuracy caused by the imbalance of the category distribution of data within the project. In this paper, in order to improve the data imbalance in the project, we use SMOTE algorithm to increase the minority samples (defective software modules), the ENN (Extended Nearest Neighbor Algorithm ) data cleaning algorithm is performed for the post-sampling data noise problem. The SA ( Simulated Annealing ) algorithm is used to optimize the four- layers BP neural network to establish the classification prediction model on the AEEEM database. We use cross validation to evaluate the performance of the proposed algorithm on AEEEM database. The results show that the proposed algorithm can effectively improve the performance of the model in predicting unbalanced data.