The Research Of Na(?)ve Bayes Classification Algorithm Based On Atrribute Reduction And Attribute Weighting

There are many DataMining methods to discover knowledge model, this papermainly studies Naive Bayesian classification model. On now days, Data Mining isplaying an increasingly important role, on people’s production and life, economicdevelopment and social progress,it also has a huge role.Classification problem in datamining is one of the most important issues has also been a concern of scholars fromvarious circles.Naive Bayesian classification model as the most widely knownclassification problem solutions but has one of the most fatal flaw, That is kind ofconditional independence assumption: Between the different conditions in the classattribute decision attribute known conditions are independent of each other; and theimportance of each condition attribute relative to decision attribute are considered to bethe same. While this provides great convenience in the calculation, but it is not in linewith real life. Thus resulting in many application scenarios Naive Bayes classificationmodel does not even have a good classification results or can not be applied.Based onthe rough set theory and information theory, this paper proposes a new Bayesclassification model, which can improve the classification accuracy.The follows are the main contents of this paper: First, researchs the Bayes theoryand Bayesian classification model, introducesBayes decision criteria, maximumassuming posteriori and describes the process of Bayes classification with examples.Second, we talk about the basic theory of Rough Sets, and on this basis, we proposetwo basic attribute reduction algorithms, which one is based on the positive domaindistinguish matrix reduction and the other is relative reductionbased ongreedyalgorithm. Third, we introduced concept of entropy and conditional entropy in Information Theory. And proposed several methods to calculate attribute importance,By combining these methods with the Na ve Bayes Classification model, then we canconstruct a new classification model, the Attribute Weighted Bayes Classificationmodel, The model is based on the importance of each measurement condition attributesand decision attribute their relative was given different weights in order to improve theclassification accuracy of Bayesian Classification model. After that, we present a newattribute reduction model based on Rough Set theory. Combining this new attributereduction moder with the Attribute Weighter Bayes Classification model, we get thefinally classification model, which is based on attribute reduction and attributewighting. At last, we prove the efficent of the new combined model through simulationwith UCI data sets and WEKA IDEs.