摘要
本文提出一种变精度粗糙集(Variable precision rough sets,VPRS)与概率神经网络(Probabilistic Neural Network,PNN)杂合的方法.变精度粗糙集对噪声数据有一定的相容性,给定置信阈值β,通过变精度粗糙集模型将信息系统中的冗余属性排除,求出一个最小的知识表示,由此可以约简神经网络的输入.由于概率神经网络的分类及泛化能力较强,接下来应用概率神经网络建立的模型进行分类、预测.实验表明,变精度粗糙集与概率神经网络杂合方法的分类及预测精度均较高.该方法可用于从模糊的、冗余的、不完备的且有噪声的大型数据库中发现知识.
The paper proposes a hybrid approach of VPRS （variable precision rough set） and PNN （probabilistic neural network）. Variable precision rough sets model is tolerant of noise. Given a confident threshold value β, redundant attributes are eliminated from information system, and a minimal knowledge representation is deducted through variable precision rough sets model. The elimination can remove the redundant input of network. Subsequently, the reduced information table is forwarded to probabilistic neural networks for classification and prediction. The research reveals that the approach of hybrid VPRS and PNN has a high accuracy in classification and prediction. The method can he applied to knowledge discovery from ambiguous, incomplete and noisy database.
出处
《情报学报》
CSSCI
北大核心
2005年第4期426-432,共7页
Journal of the China Society for Scientific and Technical Information
基金
国家自然科学基金，江苏省自然科学基金
