基于拟似然估计方法的软件失效预测模型

软件缺陷预测是软件可靠性研究的一个重要方向。由于影响软件失效的因素有很多,相互之间关联关系复杂,在分析建模中常用联合分布函数来描述,而实际应用中难以确定,直接影响软件失效预测。基于拟似然估计提出一种软件失效预测方法,通过主成分分析筛选影响软件失效的主要影响因素,建立多因素软件失效预测模型,利用这些影响因素的数字特征(均值函数和方差函数)以及采用拟似然估计方法估计出模型参数,进而对软件失效进行预测分析。基于两个真实数据集Eclipse JDT和Eclipse PDE,与经典Logistic回归和Probit回归预测模型进行实验对比分析,结果表明采用拟似然估计对软件缺陷预测具有可行性,且预测精度均优于这两种经典回归预测模型。

Software Failure Prediction Model Based on Quasi-likelihood Method

Software defect prediction is an important direction of software reliability research.Because there are many factors influencing the software failure,and the relationship among them is complicated,the joint distribution function is commonly used to describe the analysis model,which is difficult to be determined in the practical application.This problem may impact software defect prediction directly.In this paper,we proposed a Quasi-likelihood method (PCA-QLM),which uses PCA to select the main metrics firstly,and then build defect prediction model.In our model,we can use the mean function and variance function of dependent variable to get the estimated parameter and then predict defects.In this paper,we draw a conclusion that PCA-QLM can apply to the software failure prediction and its perfor-mance is better than other models by comparing with probit regression forecasting model and logistic regression forecasting model based on two real datasets Eclipse JDT and Eclipse PDE.