基于启发式BP算法的软件缺陷预测模型

针对软件缺陷预测时缺陷数据集中存在的类别分布不平衡问题，结合上采样算法SMOTE与Edited Nearest Neighbor (ENN) 数据清洗策略，提出了一种基于启发式BP神经网络算法的软件缺陷预测模型。模型中采用上采样算法SMOTE增加少数类样本以改善项目中的数据不平衡状况，并针对采样后数据噪声问题进行ENN数据清洗，结合基于启发式学习的模拟退火算法改进四层BP神经网络后建立分类预测模型，在AEEEM数据库上使用交叉验证对提出的方案进行性能评估，结果表明所提出的算法能够有效提高模型在预测类不平衡数据时的分类准确度。

Software Defect Prediction Model Based on Heuristic BP Algorithm

In order to solve the problem of unbalanced category distribution in the defect data set during software defect prediction,combined with the upsampling algorithm SMOTE and Edited Nearest Neighbor (ENN) data cleaning strategy,a software defect prediction model based on heuristic BP neural network algorithm is proposed.In the model,SMOTE algorithm is used to add a few samples to improve the data imbalance in the project,and the ENN algorithm is used for data cleaning for the data noise problem after sampling,combined with the simulated annealing algorithm that is based on heuristic learning to improve the four-layer BP neural network and to establish classification prediction model.The performance evaluation results of the proposed scheme using cross-validation on the AEEEM database show that the proposed algorithm can effectively improve the classification accuracy of the model when predicting unbalanced data.