面向航天型号软件缺陷预测的属性选择方法

An Attribute Selection Method for the Aerospace-Software Defect Prediction

摘要:

针对当前航天型号软件模块故障预测中度量属性过多而影响分类模型效率、准确率和数据收集困难的问题,以NASA公布的MDP数据集为基础,设计了基于Logistic回归模型的属性选择算法；通过该算法在6个数据集上的应用选择了对航天型号软件质量影响较大的4个属性,并将其在12个软件中进行验证,说明了该算法的有效性,选择的4个度量属性可为当前航天型号软件的度量提供参考和依据。

Abstract:

The excess measure properties for aerospace-software module defect prediction have affected the efficiency and accuracy of classification model and make it difficult to collect data. Based on the MDP data sets announced by NADA, an attribute selection algorithm is designed based on Logistic Regression Model. By applied to the six data sets, four attributes which have great impact on aerospace-software are selected. Then the four attributes are applied on 12 software of MDP, and the results prove the validity of the designed attribute selection algorithm, and the four attributes can provide reference and basis for the measure of the current aerospace software.