基于条件信息熵的自主式朴素贝叶斯分类算法  

Classification algorithm for self-learning Nave Bayes based on conditional information entropy

作　　者：邓维斌[1] 黄蜀江[1] 周玉敏[1] 

机构地区：[1]重庆邮电大学经济管理学院,重庆400065

出　　处：《计算机应用》2007年第4期888-891,共4页journal of Computer Applications

基　　金：重庆邮电大学自然科学基金研究资助项目(A2006-43)

摘　　要：朴素贝叶斯是一种简单而高效的分类算法,但其条件独立性和属性重要性相等的假设并不符合客观实际,这在某种程度上影响了它的分类性能。如何去除这种先验假设,根据数据本身的特点实现知识自主学习是机器学习中的一个难题。根据RoughSet的相关理论,提出了基于条件信息熵的自主式朴素贝叶斯分类方法,该方法结合了选择朴素贝叶斯和加权朴素贝叶斯的优点。通过在UCI数据集上的仿真实验,验证了该方法的有效性。Naive Bayes algorithm is an effective simple classification algorithm. But two central assumptions made by the Naive Bayes approach are that the attributes are independent within each class and the importance of the attributes is equal, which can harm the classification process to some extent. It is a very difficult problem in machine learning to carry out selflearning knowledge according to the characteristic of source data without prior domain knowledge. Based on the theory of rough set, a new Naive Bayes method named Conditional Information Entropy-based Algorithm for Self-learning Nive Bayes （CIEBASLNB） was proposed, which combined the merits of selective Naive Bayes （SNB） and Weighted Naive Bayes （ WNB）. Simulation results on a variety of UCI data sets illustrate the efficiency of this method.

关 键 词：朴素贝叶斯 粗糙集 条件信息熵 自主式学习 分类 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]