Research On Weighted Naive Bayesian Classification Algorithm Based On Rough Set Theory

As one of the results of natural evolution of information technology, data mining hascaused great concern from the entire information industry. Data mining is a process thatcollects knowledge and information which are used in decision support, informationmanagement, etc., from a large amount of raw data which include structured, semi-structured,text, graphics and images, with mathematical, non-mathematical, deductive and inductiveknowledge discovery method. And the knowledge discovery methods are implicated in theraw data. Data classification is a major aspect of data mining. The main classificationtechniques include Bayes classifier, classifier based on decision rules, decision tree classifier,rough set, genetic algorithms and Bayes belief network, etc.Rough sets was proposed by Z.Pawlak, an academician of Polish Academy of Sciences in1982. Since then, rough set theory has been widely concerned, it is now mainly used inapproximate sorting, attribute and attribute value reduction, attribute dependency analysis, etc.Rough set can keep the classification capacity unchanged in knowledge reduction, based onthis characteristic, we combine rough set theory and Naive Bayes classification algorithmwhich is widely used but limited by conditional independence assumption. We make thesetwo methods draw the strong points of each other to offset their own weaknesses, and in thisway, the classification capacity of Naive Bayes algorithm is improved. The research jobs ofthe paper mainly include two aspects:1. Research of algorithm combined with attribute reduction and Naive Bayes. We willanalyze several attribute reduction algorithms, including the algorithm of value reduction andattribute reduction. Attribute reduction algorithm based on attribute order has its advantage,we combine it with Naive Bayes classification algorithm, and based on this, we propose aweighted Naive Bayes classification algorithm based on attribute order. Based on thereduction of attribute order, we can simplify the data set on one hand, and participate in thecalculation of weights on the other hand. In this way, the advantages of Naive Bayes and rough set reduction are utilized effectively, and so the accuracy and adaptability ofclassification are improved to some extent.2. We combine the solving method of core attribute in rough set with Naive Bayesalgorithm, propose a weighted Naive Bayes classification algorithm based on core attribute.This weakens the limiting condition of conditional independence assumption of NaiveBayes. The solving method of core attribute simplifies the data set, this prepares aprecondition for the realization of the Naive Bayes algorithm, and the accuracy ofclassification is improved.