Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (18): 42-49.DOI: 10.3778/j.issn.1002-8331.1908-0015

Previous Articles     Next Articles

Novel Naïve Bayesian Incremental Learning Algorithm Based on Three-Way Decisions

HAN Suqing, CHENG Huiwen, WANG Baoli   

  1. 1.Department of Computer Science and Technology, Taiyuan Normal University, Jinzhong, Shanxi 030619, China
    2.School of Mathematics and Information Technology, Yuncheng University, Yuncheng, Shanxi 044000, China
  • Online:2020-09-15 Published:2020-09-10

三支决策朴素贝叶斯增量学习算法研究

韩素青,成慧雯,王宝丽   

  1. 1.太原师范学院 计算机科学与技术系,山西 晋中 030619
    2.运城学院 数学与信息技术学院,山西 运城 044000

Abstract:

Incremental learning is a kind of method that updating the classification model by modifying the parameters with the useful information in the incremental data. The Naïve Bayes algorithm is one of the best selections of incremental learning for its characteristics of natural utilizing the prior information and incremental information. The three-way decision is a promising theory proposed in recent years, which conforms to the human cognitive model with the own subjective characteristics. A novel Naïve Bayesian incremental learning algorithm based on the three-way decision is proposed in this paper, which merges the thought of three-way decision into the Naïve Bayesian model. A classification particular factor is firstly defined and applied to determine the positive, deferment, and negative regions by combining with the cost function. Then the three-region information is constructed the newly Naïve Bayesian incremental learning algorithm. The experimental results show that the classification accuracy and recall rate of this method is significantly improved when the thresholds α and β are determined appropriately.

Key words: three-way decision, Naïve Bayes, incremental algorithm, categorical sureness, deferment region

摘要:

增量学习利用增量数据中的有用信息通过修正分类参数来更新分类模型,而朴素贝叶斯算法具有利用先验信息以及增量信息的特性,因此朴素贝叶斯算法是增量学习算法设计的最佳选择。三支决策是一种符合人类认知模式的决策理论,具有主观的特性。将三支决策思想融入朴素贝叶斯增量学习中,提出一种基于三支决策的朴素贝叶斯增量学习算法。基于朴素贝叶斯算法构造了一个称为分类确信度的概念,结合代价函数,用以确定三支决策理论中的正域、负域和边界域。利用三个域中的有用信息构造基于三支决策的朴素贝叶斯增量学习算法。实验结果显示,在阈值[α]和[β]选择合适的情况下,基于该方法的分类准确性和召回率均有明显的提高。

关键词: 三支决策, 朴素贝叶斯, 增量算法, 分类确信度, 边界域