DATA CLASSIFICATION USING DIRICHLET MIXTURES

Abstract

In the article, we present a method for data classification that is based on the Dirichlet mixture statistics. An important property of the method is its ability to classify data of any type. To test performance of the method, we implemented it as a stand-alone program and tested it on the three different databases of real data. Receiver operating characteristics of the classification was used to compare the method of Dirichlet mixtures to the other classification methods. The classification results and its performance are discussed in the article. The practical value of this study is that the method based on the complex statistics is implemented as a tool and compiled as a library for further development of machine learning environments.