摘要
研究了如何利用神经网络解决软件关键模块的识别问题。首先利用交叉确认改进了级联相关算法,设计了多层前馈神经网络作为模式分类器,以软件模块的复杂性度量作为特征向量识别软件中的关键模块。最后以自行开发的维修性分配与预计(MAP)软件为例说明了采用改进的级联相关算法确定软件关键模块的优势。
How neural network could deal with the pattern recognition of software critical modules was studied. First, the cascaded-correlation algorithm was modified using cross validation, and then based on the complexity metrics of software modules, the multilayer neural network classifier was designed to identify critical modules of software. Finally, by analyzing the application in the project MAP, the experiment result shows the advantage of the modified cascaded-correlation algorithm.
出处
《计算机应用》
CSCD
北大核心
2005年第6期1336-1338,共3页
journal of Computer Applications
关键词
级联相关
关键模块
交叉确认
软件复杂性度量
模式识别
cascade-correlation
critical module
cross-validation
software complexity metric
pattern recognition
