Abstract: In order to overcome the defaults of the model including slow convergence and low accuracy in software quality prediction method based on Back Propagation (BP) neural network, a Software Quality Prediction method based on Ant Colony Optimization (ACO) improved BP Neural Network (SQP-ACO-BPNN) was proposed to determine the best network structure, initial connection weights and thresholds of the network before training the model in the proposed method. Firstly, the software quality evaluation factors were selected and a software quality evaluation system was considered. Secondly, BP neural network was adopted to build software quality prediction model and ACO was used to determine network structures, initial connection weights and thresholds of network. Then, an evaluation function  was given to select the best structure, initial connection weights and thresholds of the network. Finally, the network  was trained by BP algorithm, and the software quality prediction model was obtained. The experimental results of predicting the quality of airborne embedded software showthat the accuracy, precision, recall and F1 value of theoptimized neural network model with faster convergence are all improved, which indicate the validity of the proposed method.

Key words: software quality prediction, Ant Colony Algorithm (ACO), neural network, network structure evaluation, Back Propagation (BP)

摘要： 为了克服反向传播(BP)神经网络软件质量预测模型收敛慢，模型精度不高的问题，提出一种基于蚁群优化(ACO)反向传播神经网络的软件质量预测方法(SQP-ACO-BPNN)，在所提方法中的模型训练前确定最佳网络结构、网络初始连接权值和阈值。首先，选择软件质量评价指标，确立软件质量评价体系；然后，采用BP神经网络构建软件质量预测模型，并利用ACO确定若干网络结构、网络初始连接权值和阈值；接着，给出网络结构评价函数，选择神经网络模型的最佳结构、网络初始连接权值和阈值；最后，通过BP算法训练该网络，得到软件质量预测模型。在机载嵌入式软件质量预测数据上的实验结果表明，优化后的神经网络模型有效提高了准确率、精确率、召回率和F1值，并且能够更快收敛，验证了所提方法的有效性。

关键词: 软件质量预测, 蚁群算法, 神经网络, 网络结构评价, 反向传播