Study on Models and Methods of Pattern Recognition of Bayesian Network Based on Rough Set

Abstract: The pattern recognition method of Bayesian networks has a soft and fault-tolerant merit and can deal with the recognition problem of incomplete information very well, it is one of most effective theory models in expression of uncertainty knowledge and it becomes one of the fields in data mining and pattern recognition which studied most. In this paper, on the basis of analyzing the developing process and current research situation of Bayesian networks, Bayesian classifier and its applications in pattern recognition, the learning algorithm of structures and parameters of Bayesian network is researched to overcome the difficulty in this problem, the recognition models and methods of Bayesian network based on rough set are put forward and its feasibility and validity are verified by experiments. The detailed information is described below:(1) On the basis of the classic Pawlak rough set theory, the rough set model of general relationship and the variable precision rough set model, the generalized variable precision rough set model is proposed to raise the efficiency of attribution reduction and the ability of anti-jamming of Bayesian network classifier for noise datum;(2) Theβ-reduction algorithm of the variable precision rough set of complete information system and theα~βlower approximation reduction algorithm of the variable precision rough set of incomplete information system are established to overcome the NP-hard problem of seeking minimum reduction by the definition of the generalized variable precision rough set;(3) The computational complexity of Bayesian network is reduced by attribution reduction and extracting decision rules, and the structures of Bayesian network is trained by utilizing the information of certainty factor and coverage factor, at last the parameters of Bayesian network is learning on the basis of Bayesian network structures and the statistics information of attribution value in decision rules and Bayesian network based on rough set is acquired;(4) The minimum error rate Bayesian decision-making criteria based on rough set is put forward to create Bayesian network classifier based on rough set by using rough set Bayesian formula;(5) A comprehensive comparison of RBAN classifier, Na?ve Bayes (NB) classifier and BN Augmented Naive-Bayes (BAN) classifier is done by experiments. Whether complete information system, or incomplete information system, the result shows that RBAN classifier has most accuracy, its speed is slightly slower than NB classifier and significantly faster than BAN classifier…
Key words: Rough Set; Bayesian Network; Classifier; Pattern Recognition

This entry was posted in Master Thesis. Bookmark the permalink.