Using Bayesian Classifiers to Enhance Clustering

Recently, combining Naive-Bayes with the expectation maximization (EM) algorithm for unsupervised learning have received significant attention. AutoClass is a classical Bayesian clustering algorithm that uses Naive-Bayes in combination with EM algorithm to find the probability distribution parameters to best fit the data. In this study, we introduce a robust approach, which is similar to AutoClass, it can arbitrarily impose any Bayesian classifiers in combination with EM algorithm to enhance cluster´s performance. This paper focuses on how clustering techniques can benefit from classification. We provide experimental evidence that more accurate than original results of clustering in the t-test on most of the benchmark data sets