Learning Huge Bayesian Network Classifier with Augmented Naive Bayes

• KIKUYA Naruchika

  The University of Electro-Communications

• SUGAHARA Shouta

  The University of Electro-Communications

• NATORI Kazuki

  The University of Electro-Communications

• UENO Maomi

  The University of Electro-Communications

Bibliographic Information

Other Title

• Augmented Naive Bayesによる大規模ベイジアンネットワーク分類器学習

Abstract

A Bayesian network classifier (BNC) is known as a probabilistic classifier for discrete variables. Previous research indicates the exact learning of BNC with the Augmented Nai¨ve Bayes (ANB) structure constraint outperforms the approximation methods. However, the exact learning cannot learn the huge BNC. On the other hand, there is a constraint-based algorithm that can learn 3500 nodes Bayesian networks by the RAI algorithm with the transitivity. This paper proposes a method that extends this algorithm for the ANB learning, and proves that the proposed method has an asymptotic consistency for ANB structures. The experimental results show that the proposed method outperforms the other methods for the huge BNC learning.

Journal

• 電子情報通信学会論文誌D 情報・システム

  電子情報通信学会論文誌D 情報・システム J104-D (1), 65-81, 2021-01-01

  The Institute of Electronics, Information and Communication Engineers

Keywords

• Bayesian networks
• probabilistic graphical models
• machine learnings
• classifiers
• constraint-based approach

Details

• CRID: 1390005506386776192
• NII Article ID: 120007188399
• DOI: 10.14923/transinfj.2020jdp7008
• ISSN: 18810225
• ISSN: 18804535
• Web Site: https://uec.repo.nii.ac.jp/records/10082
• Text Lang: ja
• Data Source:
  • JaLC
  • IRDB
  • CiNii Articles
  • KAKEN
• Abstract License Flag: Disallowed