Tree Augmented Naive Bayes Classifier Based On Attributes Reduction Using Association Rules And Its Applications

After entering the21st century, the economic and cultural community has been developingrapidly, people’s living standards have been gradually improved. People around the world areincreasingly concerned about health issues began to themselves and their families, which intoday’s medical field has put forward higher requirements. Meanwhile, with the developmentof excellence in the field of artificial intelligence and medical technology continues to progressin many areas under the trend of technology integration, computer-aided medical diagnostictechnology to become one of the many scholars who research focus.Bayesian network is a common knowledge representation and reasoning tools in one, is animportant area of research in artificial intelligence, has been studied by many scholars all overthe world and is widely used in various fields. Bayesian network is a directed acyclic graph,combined probability theory to express the causal relationships between variables. Bayesiannetworks have a simple structure and a complete clear semantics, its auxiliary medicaldiagnosis also has a wide range of applications. However, when studying the structure of theBayesian network, the number of sets of attribute data, the greater the complexity of thenetwork structure learning, will get more complicated network structure. And construct thebest fit with the given data network structure is a NP-HARD problem. If we can make theappropriate dataset attribute reduction, it can be simplified Bayesian network structure, whichhas important significance for the auxiliary medical diagnosis, that uses less information andtime to get a high diagnostic accuracy.In this paper is presented based on association rules attribute reduction and dynamicselection based on specific attributes of the training set of Bayesian methods AD-TAN (theAssociation rules and Dynamic training set based Tree Augmented Naive bayes). The methodto be determined according to the classification of each instance of the dynamic characteristicsof the training set for instance, by removing important is the lack of missing values and theircorresponding attributes to complete. By demonstrating the association rules in the databasefunctional dependency discovery, association rules, in some cases in the database have acommon characteristic functional dependency. These features will be found in the relationshipbetween attributes, you can find out the attributes that can be removed. By association rulemining, mining association rules discovery focus, the right piece of property that can be usedin a rule attribute reduction. So based on dynamic training set selection and TAN associationrules attribute reduction based approach, namely AD-TAN methods.Since the TAN tree using AD-TAN association rules to remove redundant attributes, in fact the tree is a node on a path to build TAN deleted when the network structure, while the othernodes on the path unchanged. Through the study of probability and derivation rules, drawnimportant conclusions can be used for attribute reduction, which is based on the derivation ofArmstrong axioms. These conclusions can be found by those who come through redundantassociation rule mining properties, is a path among the Bayesian network structure, startingfrom the property until the leaf nodes of the path of these nodes can be used as a redundantnode. Finally, the conclusions derived probability rules and association rules combiningcharacteristics to identify those nodes and their associated rules noted in all descendants of thenode tree structure as a possible redundant nodes. These may be redundant nodes may have animpact on the accuracy of the classification, the paper layer by layer through iterative methodto remove a node, and the introduction of the parameter to be controlled. Based on this proposedIC-TAN method (Iterative Control Method Based on AD-TAN with adding and deletingAttributes, attribute-based iterative control method with the addition and subtraction of AD-TAN).IC-TAN method can be applied in the auxiliary field of medical diagnosis using the method,the physician may in a short time and using less information for patients for disease diagnosis.Use IC-TAN method for each instance of the specific characteristics of the case, plus a goodmatch with the medical diagnosis. Using this method can help doctors diagnose quickly andaccurately.In order to verify IC-TAN classification results, the paper experiment using16UCI datasets, and with the NB,TAN,SNB-CFS and TAN-IGS made precision contrast. Experimentsshow that, IC-TAN has a better classification results, and requires less information.