Software defects prediction can help raise the effectiveness and efficiency of testing activities by constructing predictive classification models from static code attributes which can identify software modules with a higher than usual probability of defects. Our aim is to find the best performance predictive classification model through introducing SVM into DP. Sections 1 through 4 of the full paper explain our SVM-DP model and its application to analyzing the 13 data sets of NASA Metrics Data Program (MDP). Sections 1 through 4 are entitled; Iterative and Incremental Prediction Model SVM-DP ( section 1) ; Benchmarking Data Sets and Code Metrics ( section 2 ) ; Effectiveness Indicators ( section 3 ) ; Experimental Method and Analysis of Test Results ( section 4). Experimental results , presented in Table 4 and Figs. 4 through 7, and their analysis, show preliminarily the effectiveness of our SVM-DP model.%软件缺陷预测在软件系统开发的各个阶段发挥着极为重要的作用.利用机器学习的相关方法建立更好的预测模型已经被广泛研究.文章分析了支持向量机SVM作为二值分类模型应用到软件缺陷预测中的实现方法,构造了基于SVM的可迭代增强的缺陷预测模型SVM-DP.在13个基准数据集上开展比较实验,定量地分析了应用各种核函数对SVM-DP模型性能的影响.实验结果显示,应用线性内积核函数的SVM-DP具有最优的预测性能.同时,在与J48的比较实验中,最高超过J48预测模型20%的性能进一步证明了SVM-DP模型应用于软件缺陷预测的有效性.
展开▼