摘要
针对在小样本情况下BP神经网络在生成软件测试用例的过程中可能产生的过学习问题及识别正确率较差的缺点,运用支持向量机具有更好的泛化性能的原理,提出了应用支持向量机生成测试用例的方法。对五个软件测试实例针对多组不同数量的训练样本所做的实验表明,在小样本情况下与BP神经网络相比,应用支持向量机得到的测试用例预期结果的正确率提高了10个百分点以上,说明了该方法的有效性。
In the condition of small samples, overlearning problems and the disadvantage of poor recognition accuracy may oc- cur in the process of generating software test cases by BP neural network. As SVM has better generalization performance, this paper proposed a method of using SVM to generate test cases, and performed experiment on five practical cases with multi- groups of different number of training samples. The results show that, in the condition of small samples, compare with BP neu- ral network, this method can improve the accuracy of expected results by more than 10 percentage points. The results indicate the effectiveness of this method.
出处
《计算机应用研究》
CSCD
北大核心
2015年第1期115-120,共6页
Application Research of Computers
基金
江苏省产学研项目(BY2013015-40)