Research And Application Of Several Improved Naive Bayesian Classification Algorithms

Classification as the core research content in the field of data mining has a very wide range of applications in real life.For example,according to the patient's clinical illness to determine what the patient is suffering from.There are many common methods for constructing classifiers,such as bayesian networks,support vector machines,artificial neural networks,and fuzzy sets and any more.As a classical classification algorithm in the family of bayesian algorithm,naive bayes has attracted the attention of the majority of scholars,because of its simple structure and high computational efficiency.The naive bayes classification algorithm is based on the assumption that when a class tag is given,the attribute values are mutually independent.Although this kind of assumption makes its computation simple,it also limits its classification performance on the data sets with strong correlation between many attributes.Therefore,scholars have put forward many improved algorithms through relaxing its assumption.The averaged one-dependence estimators and hidden naive bayes are very good improved naive bayes classification algorithms,which not only greatly improve the classification accuracy of the original algorithm,but also have a good performance on many different kinds of data sets.In this paper,the averaged one-dependence estimators and hidden naive bayes are taken as the basic research objects.Considering the influence of data types on the classification in practical applications,we propose two averaged one-dependence estimators based on attribute weighting and a hidden naive bayes based on attribute value weighting.The specific research content is as follows:(1)By studying the naive bayes classification and the averaged one-dependence estimators,the averaged one-dependence estimators based on Tau-y and Lambda-y is proposed respectively,and two improvements are verified by numerical experiments.The classification performance of the two improved algorithm has been significantly improved compared to the original algorithm.(2)By studying the naive Bayes classification and hidden naive bayes,through integrating the statistical information formulas of the related attributes,attribute values,and class labels that was used when the classifier of hidden naive bayes was constructed,proposed a hidden naive bayes based on attribute values.And numerical experiments show that the classification performance of the improved algorithm is significantly improved compared with the original algorithm.(3)With the time complexity,classification accuracy and AUC value as indicators,the advantages and disadvantages of the three improved algorithms proposed in this paper are compared and their future research directions are proposed.(4)Applying the three improved algorithms proposed in this paper to the field of TCM diagnosis of asthma and building a whole set of modeling methods for TCM diagnosis of asthma,and then comparing the performance of the improved algorithms and the original algorithms through numerical experiments,further test the effectiveness of the improved algorithms.