维规约对朴素贝叶斯分类性能的影响研究

ON EFFECT OF DIMENSION REDUCTION ON CLASSIFICATION PERFORMANCE OF NAVE BAYES

朴素贝叶斯算法是一种简单而高效的分类算法,但属性的条件独立性假设并不符合客观实际,特别是高维度数据的属性之间往往存在相关关系,如何能在实现对数据降维的同时又提高朴素贝叶斯的分类性能是一个重要的研究问题。对基于条件信息熵的选择朴素贝叶斯、基于主成分分析的朴素贝叶斯和基于独立成分分析的朴素贝叶斯算法进行研究,通过在UCI数据集上的仿真实验,详细比较了几种维规约算法对朴素贝叶斯分类性能的影响。

Nave Bayes（NB） algorithm is an effective and simple classification algorithm,but its conditional independence assumption is not in compliance with the objective reality,especially the attributes in high-dimension data set are always dependent on each other.It is an important research subject to reduce the dimensions of the data while improve the classification performance.In the paper,conditional information entropy-based selected nave Bayes（CIESNB）,principal components analysis-based na？ve Bayes（PCANB） and independent components analysis-based nave Bayes（PCANB） are studied.The effects of these dimension reduction algorithms on NB classification performance are compared in detail through the simulation experiment carried out on UCI data sets.