基于ICA与Bayes的判别分析模型  

Discriminatory analysis model based on ICA and Bayes

作　　者：彭红毅[1] 蒋春福[2] 朱思铭[3] 

机构地区：[1]华南农业大学理学院,广州510642 [2]深圳大学数学与计算科学学院,广东深圳518060 [3]中山大学数学与计算科学学院,广州510275

出　　处：《计算机应用研究》2007年第8期58-59,96,共3页Application Research of Computers

基　　金：国家自然科学基金资助项目(10371135)

摘　　要：简要介绍了Bayes判别分析模型的特点及存在的问题,概括了独立成分分析(ICA)的特点及发展现状,提出了基于ICA与Bayes的判别分析模型——IBD模型。该模型首先利用ICA的方法将相关性数据指标转换为互相独立的数据指标,并通过卡尔曼滤波方式滤去高频数据,有效地去除了噪声,最后利用Bayes方法对转换的数据进行判别分析。实验结果表明,当数据之间存在相关关系时,IBD模型的判别分析效果要优于Bayes与Fisher判别分析模型。This paper introduced the characteristics and defects of Bayes discriminatory analysis model, summarized the ICA＇ s characteristics and development state. Then based on ICA and Bayes discriminatory analysis model, it provided an IBD model. Firstly, IBD model transformed correlative data indicators into mutually independent data indicators, then f＇dtered of the high-frequency data by ICA, so that it got rid of the noise effectively. At last, discriminated the data to a cluster by Bayes. The experiments show that the discriminatory effect of IBD is prior to that of Bayes and Fisher in dealing in the correlative data.

关 键 词：独立成分分析 贝叶斯 相关关系 

分 类 号：O212[理学—概率论与数理统计] TP301.6[理学—数学]