计算机软件质量综合评判的人工神经网络模型

根据单指标评价标准，通过在规定的区间范围内随机取值生成足够多的训练样本、检验样本和测试样本，遵循误差反传（BP）网络建模基本原则和步骤，建立可靠、有效的软件质量综合评价模型，克服了物元模型、灰色模型等方法存在受人为因素影响及定性评价等缺陷．对6个模拟样本和8个实际软件的评价表明，本文提出的样本生成方法、建模过程能有效避免出现过训练和过拟合现象，建立的BP模型具有较好的泛化能力、可比性和客观性；各评价指标与软件质量等级之间存在明显的非线性关系，可靠性指标对软件质量等影响最大．

Comprehensive evaluation model for computer software quality using artificial neural network

The sample data for training,verification and testing,in agreement with the singular evaluation index(EI) criterion,were randomly generated.The artificial neural network(ANN) based model with reliability and effectiveness for computer software quality(CSQ) evaluation was established in accordance with the basic principles and steps.The case study shows that the methodology for generating sample data and the process of establishing ANN-model are effective and reliable.The phenomenon such as over-training and...