基于代价敏感半监督的跨项目软件缺陷数预测模型  

Cross-project Software Defect Number Prediction Model Based on Cost-sensitive Semi-supervised

作　　者：高晶[1] GAO Jing(Department of Information Engineering,Yantai Vocational College,Yantai 264670,Shandong Province)

机构地区：[1]烟台职业学院信息工程系,山东烟台264670

出　　处：《沈阳工程学院学报（自然科学版）》2022年第1期75-80,共6页Journal of Shenyang Institute of Engineering:Natural Science

摘　　要：软件日益推陈出新,当前方法无法满足源项目与目标项目的预测需求,因此创建一种基于代价敏感半监督的跨项目软件缺陷数预测模型。首先,利用斯皮尔曼等级相关系数选择方法,选取合适度量元,按照系数降序对其实施排列,得到所需的度量元集合。其次,运用版本控制工具,完成版本源代码收集,从而获取缺陷数据。再次,经过静态分析统计源代码,架构度量元矩阵,并通过引入极小值方法,实现属性值对数化预处理,再基于源代码度量元与缺陷之间的潜在相关性,对候选集样本实施采样。然后,寻求一个代价敏感半监督支持向量机的超平面,使最小化分类整体代价得以实现。最后,根据预测模型的设定参数与训练学习,完成预测模型整体设计。仿真实验表明,该模型不仅能够有效预测多种形式的跨项目缺陷数,而且精准率与错分率均具有明显的优越性,效果十分理想。Software is becoming more and more innovative,and the current method cannot meet the forecasting needs of the source and target projects.Therefore,this paper creates cost-sensitive semi-supervised cross-project software defect number prediction model,which uses Spearman rank correlation coefficient selection method,and selects suitable metrics.The elements are arranged in descending order of coefficients to obtain the required set of measurement elements.Version control tools are used to complete the collection of version source code to obtain defect data.After static analysis of the source code,the measurement element matrix is constructed,and a small value method,which implements logarithmic pre processing of attribute values,based on the potential correlation between source code measurement element sand defects,samples the candidate set samples,and then seeks a hyper plane of cost-sensitive semi-supervised support vector machines to minimize The overall cost of classification is realized.Finally,the overall design of the prediction model is completed according to the set parameters and training of the prediction model.Simulation experiments show that the model in this paper cannot only effectively predict the number of cross-project defects in various forms,but also have obvious advantages inaccuracy and misclassification rate,and the effect is very satisfactory.

关 键 词：代价敏感 半监督 跨项目 软件缺陷数预测 度量元 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]