Session:

Knowledge Representation, Decision Support and Expert Systems
Tuesday, March 02, 2004, 17.15 – 17.35

Session Chair:
Vice Chair:

A. Dobnikar
M. Savoji

Paper Title:

Local Estimate of Distribution Mapping Exponent for Classification of Multivariade Data

Author(s):

M. Jirina, Institute of Computer Science AS CR, Czech Republic

Abstract:

Methods for classification of multivariate data which are based on the nearest neighbors approach solve the problem of classification by an estimate of the probability density at point x of the data space by ratio i/V. i is the number points of a given class of the training set in a suitable ball of volume V with center at point x. The method proposed is based on notion of distribution mapping exponent and its local estimate q for each point x. Distances of all points of a given class of the training set from a given (unknown) point x are used for the probability density estimate. It is shown that the sum of reciprocals of q-th power of these distances can be used as the probability density estimate. The classification quality was tested and compared with other methods using multivariate data from UCI Machine Learning Repository. The method has no tuning parameters.