基于改进ACO优化BPNN的软件缺陷预测模型

Optimizing software defect prediction model of BP neural network based on improved ACO algorithm

针对用BP神经网络进行软件缺陷预测时出现的易陷入局部最优、学习速度缓慢等缺陷问题,提出一种基于信息素初始化和局部路径优化的蚁群优化算法优化BP神经网络的软件缺陷预测模型。对待预测的数据集进行基于互信息和自信息优化的主成分分析操作,降低数据的维数,提高运算效率;根据改进后的蚁群优化算法,计算最优的BP神经网络权值和阈值;使用NASA提供的软件缺陷数据集,利用提出的模型进行缺陷预测,基于十折交叉方法进行验证。通过与几种传统方法对比验证了所提方法具有更快的收敛速度和更高的预测准确度。

The traditional BP neural network algorithm has many drawbacks when it is used in software defects prediction. For example, it falls into local optimum easily and learns slowly. In view of these problems, a software defects prediction model based on the BP neural network improved by ant colony optimization algorithm （ACO） was proposed. Specifically, the ACO algorithm was based on pheromone initialization and local path optimization. The mutual information and self-information were integrated with the principal component analysis （PCA） method to reduce the dimensions of data set and increase the operation efficiency. The optimal BP neural network weights and thresholds were calculated using the optimized ant colony optimization algorithm. The performances of the proposed model were tested on NASA data sets according to the ten-fold cross method. Experimental results show that the proposed method achieves higher convergence rate and accuracy compared with conventional methods.