Learning distributed bayesian network structure using majority-based method

Article type: Research Article

Authors: Shetty, Sachin^a | Song, Min^{b; *} | Yang, Houjun^c | Matthews, Lisa^b

Affiliations: [a] Department of Electrical and Computer Engineering, Rowan University, Glassboro, NJ 08028, USA | [b] Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA 23529, USA | [c] Department of Computer Science, Qingdao University, Qingdao, China

Correspondence: [*] Corresponding author. Tel.: +1 757 683 5194; Fax: +1 757 683 3220; E-mail: msong@odu.edu.

Abstract: In this paper we present a majority-based method to learn Bayesian network structure from databases distributed over a peer-to-peer network. The method consists of a majority learning algorithm and a majority consensus protocol. The majority learning algorithm discovers the local Bayesian network structure based on the local database and updates the structure once new edges are learnt from neighboring nodes. The majority consensus protocol is responsible for the exchange of the local Bayesian networks between neighboring nodes. The protocol and algorithm are executed in tandem on each node. They perform their operations asynchronously and exhibit local communications. Simulation results verify that all new edges, except for edges with confidence levels close to the confidence threshold, can be discovered by exchange of messages with a small number of neighboring nodes.

Keywords: Bayesian network, distributed data mining, peer-to-peer networks, majority voting

DOI: 10.3233/JCM-2009-0235

Journal: Journal of Computational Methods in Sciences and Engineering, vol. 9, no. s1, pp. S53-S68, 2009