改进的软件缺陷预测模型研究

Research on improved software defect prediction model

为了提高软件缺陷预测模型的准确性,利用改进的频繁项集挖掘算法(IMMFIA)的低时间开销特点和改进的支持向量机(EDSVM)的优化能力提出改进的软件缺陷预测模型(FREDSVM)。利用IMMFIA获取频繁项集,并产生满足置信度和支持度阈值的关联规则;根据相关度和新的规则排序度量提高小类(带缺陷的软件模块)的优先级,得到分类器;运用EDSVM针对规则匹配无果问题和规则匹配溢出问题进行分类。实验证明:与当前的软件缺陷预测方法相比,FREDSVM方法具有较高的准确性。

In order to improve the accuracy of the software defect prediction model, an improved software defect prediction model (FREDSVM) is proposed with using the improved frequent item set mining algorithm (IMMFIA) and improved support vector machine (EDSVM). We use IMMFIA to get frequent item sets and generate association rules that satisfy the confidence and support thresholds. The priorities of small classes (defective software modules)based on relevance and new rules are improved and the classifier is obtained. The EDSVM is used to classify rule mismatching problem and rule matching overflow problem. The experimental results show thatFREDAVM has high precision compared with the current software defect prediction method.