摘要
孤立点挖掘是数据挖掘的一个重要领域,而统计分析方法在孤立点检测中具有天然的优势。本文将统计聚类方法融入RBF神经网络,提出了一种基于统计聚类RBF神经网络的新的孤立点检测算法—SCRBF。该算法包括两部分,先用统计聚类方法对神经网络进行初始化,然后根据网络的训练情况进行隐单元的简化,提高了神经网络的泛化能力,同时也降低了过拟合现象的出现概率。与LSC算法的对比实验表明,该算法是有效的。
Mining isolated point is an important field in Data Mining. Methods of statistical analysis have natural advantage in detecting isolated points. In this paper, statistical clustering is first integrated into RBF Neural Network and a new isolated point detecting algorithm based on statistical clustering RBF Neural Network, SCRRBF is proposed, which has two steps. The first step is initializing the neural network using statistical clustering, and the second is to reduce the concealing units of neural network according to the training situation. Using this, the generalization of neural network can be improved and the Over-fitting phenomenon can be reduced. With experimental contract to LSC algorithm, SCRBF is effective.
出处
《计算机科学》
CSCD
北大核心
2006年第10期196-197,271,共3页
Computer Science
关键词
统计方法
聚类
RBF神经网络
孤立点检测
Statistical method, Clustering, RBF neural network, Isolated point detecting
