基于支持向量回归的软件缺陷密度预测模型

Prediction Model of Software Defect Density Based on Support Vector Regression

针对软件缺陷密度的预测问题,构建了一种基于支持向量回归(Support Vector Regression,SVR)的软件缺陷密度预测模型,指出影响回归预测精度的主要因素。首先,对软件度量元数据进行提取,利用归一化和随机序列的方法对缺陷数据进行预处理,并将数据分成训练集和测试集进行回归预测;然后,引入网格搜索的方法对支持向量回归模型中的参数进行优化,大大提高了预测的精度;最后,通过实验对比其他5种机器学习算法,验证了预测模型的有效性。

In order to predict the software defect density,a prediction model of software defect density based on Support Vector Regression( SVR) is constructed. The main factors that affect SVR prediction accuracy are data processing and parameter selection. Firstly,the metric metadata of software is extracted,and the normalization and the random sequences methods are used to preprocess the defect data. The data is divided into training set and test set for regression prediction. Then the grid search method is used to optimize the parameters of the SVR model,greatly improving the prediction accuracy. Finally,the validity of the prediction model is verified by comparing experiments with the other five kinds of machine learning algorithms.