嵌入属性加权的实例加权朴素贝叶斯算法  

Embedded-Attribute Weighted Instance-Weighted Naive Bayes

作　　者：杨柳 胡桂开 彭萍 曾嘉琪 

机构地区：[1]东华理工大学理学院,江西 南昌

出　　处：《应用数学进展》2023年第5期2392-2401,共10页Advances in Applied Mathematics

摘　　要：朴素贝叶斯是一类应用广泛的分类算法,它是根据贝叶斯定理和属性条件独立来实现的。然而,属性条件独立性假设在现实生活中难以满足,为减少该假设对朴素贝叶斯算法效果的影响,本文提出了一种将属性加权嵌入到实例加权过程中的朴素贝叶斯算法。首先,基于相关性属性加权算法计算各个属性的权重;其次,将实例众数与训练实例的相似度进行属性加权,并按照不同实例众数对加权后的相似度进行算术平均得到实例权重;然后,利用实例权重构建加权朴素贝叶斯分类器;最后,采用标准UCI数据集将我们提出的算法和朴素贝叶斯算法、实例加权朴素贝叶斯算法进行仿真实验,结果表明我们提出的算法在准确率以及F1值上优于其它两种算法。Naive Bayes is a widely used classification algorithm, which is independently implemented based on Bayesian theorem and attribute conditions. However, the assumption of attribute conditional independence is difficult to meet in real life. To reduce the impact of this assumption on the per-formance of naïve Bayesian algorithms, we propose a naive Bayes algorithm by embedding attrib-ute weighting into instance weighting process. Firstly, the weight of each attribute is calculated based on the correlation attribute weighting algorithm. Secondly, the similarity between the in-stance mode and the training instance is weighted by attribute, and the weighted similarity is arithmetically averaged according to the different mode instances to get the instance weight. Then, a weighted naive Bayes classifier is constructed using case weights. Finally, the standard UCI data set is used to simulate the proposed algorithm, naive Bayes algorithm and case weighted naive Bayes algorithm. The results show that the proposed algorithm is superior to the other two algo-rithms in accuracy and F1 value.

关 键 词：朴素贝叶斯 实例加权 属性加权 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]