Classification Based on Multiple-Resolution Data View

Abstract. We examine efficacy of a classifier based on average of kernel density estimators; each estimator corresponds to a different data "resolution". Parameters of the estimators are adjusted to minimize the classification error. We propose properties of the data for which our algorithm should yield better results than the basic version of the method. Next, we generate data with postulated properties and conduct numerical experiments. Analysis of the results shows potential advantage of the new algorithm when compared with the baseline classifier. Key words: kernel density estimation, classification based on density estimation, average of density estimators

• Type: Conference
• Year: 2010
• Where: ICANN
• Authors: Mateusz Kobos, Jacek Mandziuk
• Keywords: Density Estimation, Density Estimators, ICANN 2010, Kernel Density Estimators, Neural Networks