基于模糊多目标线性规划的软件缺陷预测方法研究  

Software Defect Prediction Method Based onFuzzy Multi-criteria Linear Programming

作　　者：吴瑞霞 张志旺[1] 王琰 周莉[1] 岳峻[1] 卢泰然 WU Ruixia;ZHANG Zhiwang;WANG Yan;ZHOU Li;YUE Jun;LU Tairan(School of Information and Electrical Engineering,Ludong University,Yantai 264039,China;School of Information Science and Electrical Engineering(School of Artificial Intelligence),Jinan 250357,China)

机构地区：[1]鲁东大学信息与电气工程学院,山东烟台264039 [2]山东交通学院信息科学与电气工程学院(人工智能学院),济南250357

出　　处：《鲁东大学学报（自然科学版）》2021年第2期131-138,共8页Journal of Ludong University:Natural Science Edition

基　　金：国家自然科学基金(61877061,61872170);山东省自然科学基金(ZR2016FM15)。

摘　　要：为了提高软件开发过程的可测性和可信性,本文在分析软件缺陷预测数据特点的基础上,提出了一种新的带特征选择的模糊多目标线性规划分类器FMCLPC-FS。首先,定义了一个模糊隶属度函数来处理原始数据中的噪声和异常值;然后,利用核函数将非线性可分问题转化为线性可分或近似线性可分问题。此外,在多目标线性规划分类器MCLPC中引入了稀疏化函数,可以在分类过程中去除数据集中的冗余特征并选择出最少的重要特征。实验结果显示,与MCLPC和SVC相比,FMCLPC-FS可以显著提高缺陷预测的准确性和分类的可解释性。In order to improve the testability and credibility of software development process,a new fuzzy multi-criteria linear programming classifier with feature selection FMCLPC-FS was proposed based on the analysis of the characteristics of software defect prediction data.First,a fuzzy membership degree function was defined to deal with the noise and outliers in the original data.Then,the nonlinear separable problem was transformed into the linearly separable or approximate linearly separable problem by using kernel function.In addition,a sparse function was introduced into the MCLPC model to remove redundant features from the dataset and select the least number of important features in the classification process.The experimental results show that compared with MCLPC and SVC,FMCLPC-FS can significantly improve the accuracy of defect prediction and the interpretability of classification.

关 键 词：模糊隶属度 多目标线性规划 软件缺陷预测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]