一种软件工作量估算的不确定性度量方法  

Uncertainty Measurement Method of Software Effort Estimation

作　　者：解浪[1,2] 杨叶[2] 

机构地区：[1]中国科学院研究生院,北京100190 [2]中国科学院软件研究所互联网软件技术实验室,北京100190

出　　处：《计算机工程》2012年第3期39-42,共4页Computer Engineering

基　　金：国家自然科学基金资助项目(90718042);广东省中国科学院全面战略合作基金资助项目(2009B091300131)

摘　　要：以COCOMO81模型为基础,结合模型输出方差以及模型与数据自身方差的组合,分别度量模型本身估算值与实际值差距的不确定性,并在此基础上给出预测区间。提出以命中率(SR)与平均相对宽度(MRV)相结合的评测标准。通过采用重采样获得N对训练和测试集合,计算不同置信度下区间SR和MRV的均值,运用SR和MRV的散点图比较不同度量方法获得区间。实验结果表明,该方法能以相同的命中率获得更窄的区间。This paper proposes a methtnt to assess the uncertainty of using COCOMO81 model. It uses the variance of model＇s output to assess the uncertainty of model＇s output. And uses both the data＇s variance and the model＇s output to assess the uncertainty of the distance between estimated effort and its real value. Based on the uncertainty, it proposes the prediction interval. It proposes the Mean Relative Variance （MRV） and shooting ration as the measurement to assess the umcertainty. In the experiment, re-Sampling method is employed to get N pairs of train set and test set. The average value of SR and MRV for different confidence level is calculated. After that, the scatter plot of SR and MRV is used to compare the different intervals that generated by different methods of assessing uncertainty. Experimental result shows that the measurement for uncertainty can generate narrower interval at the context of the same shooting ratio.

关 键 词：不确定性 软件工作量估算 重采样 预测区间 COCOMO模型 方差 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]