The Hybrid Classification using Empirical Bayes and Nearest Neighbor with Stable Distribution

Abstract

The propose of this research aimed to investigate a classification technique using Empirical Bayes in combination with Nearest Neighbor (EBNN) when data are distributed as Stable-normal, Cauchy and Levy distributions. The study is performed using informative priors, normal distributions with unknown mean but known variance. Data employed in this study were generated into two equal sets, consisting of training set and test set with the sample sizes 100 and 500 for the binary classification. In each situation, the data are simulated with Monte Carlo technique and repeated 5,000 times. The average percentage of correct classification is used as criteria for comparison. The results found that EBNN method exhibited an improved performance over Empirical Bayes method in all distributions under study. Increasing neighborhood (k) exhibited higher percentages of correct classification in all distributions. Stablenormal and Levy distribution with increasing sample size also exhibited higher percentages of
correct classification.