基于ICA的多维贝叶斯分类器

多维贝叶斯分类器是处理多维分类问题的概率图形模型,其中属性变量可决定一个或多个类变量。文中针对属性变量维数较高和信息冗余问题,采用Fast ICA算法对属性变量进行降维,从而将高维属性变量约减为能较完整描述数据信息的低维属性变量。然后根据约减后的属性变量构建多维贝叶斯分类器;最终,通过理论分析得到基于ICA的多维贝叶斯分类器的性能较好。实验结果表明,对3组基准数据集的分类,基于ICA的多维贝叶斯分类器相比于其他算法具有较高的分类准确率。

Multi-dimensional Bayesian Network Classifiers Based on ICA

Multi-dimensional Bayesian network classifiers （MBCs） are probabilistic graphical models proposed to deal with multi-dimensional classification problems, where each feature variable determines one or more than one class variable. For the problem of high dimensional of feature attributes and information redundancy, the Independ- ent Component Analysis （ICA） is applied to decrease the dimension of feature variable which could completely de- scribe data. Then, we construct a multi-dimensional Bayesian network classifier according to the decreased data. Finally, the performance of the MBCs is proved good by theoretical analysis. The experiment results show that for three benchmark multi-dimensional data sets, the multi-dimensional Bayesian network classifier based on ICA outper- forms other algorithms in accuracy.