Ordering and Elimination Based Component Learning Method

In this paper, we propose a component learning method to learn a set of Gaussian components that fit the given data distribution. An ordering and visualization technique called OPTICS and tests of multi normality are used in this method. We consider the applications of the proposed method to the tasks of classification and clustering. Here, the components are used to define a feature space to which the data points are transformed. In that feature space, classification is performed using linear support vector machines and clustering is performed using support vector clustering. The performance of the component learning method and its application to classification and clustering is demonstrated on synthetic datasets.