A comparison of parametric and non-parametric techniques for pattern classification on Canadian agricultural loan data

Abstract

Pattern classification involves learning a model from a set of labeled training samples that can in turn be used to help determine the label of new samples encountered. The models themselves take on one of two forms: parametric models which make assumptions about the form of the distribution of sample features given a label; or non-parametric models which make no assumptions about the form of the distribution but retain more knowledge of the training samples to assist with labeling new objects. In this paper we consider data on Canadian Agricultural Loans and show that despite the fact that the underlying data does not fit the assumptions made by the parametric techniques they still perform as well or better than the non-parametric techniques.