The Performance of Classification Using Empirical Bayes in Heavy-tailed Distribution
N. Deetae, Saowanit Sukparungsee, Yupaporn Areepong, Katechan Jampachaisri · International Journal of Applied Mathematics & Statistics/International journal of applied mathematics and statistics · 2013
The objective of this study is to develop a classification technique in normal and heavy-tailed asymmetric distributions, using Empirical Bayes method (EB) with conjugate priors, normal distributions with known mean but unknown variance. The results are compared to Classical and K-nearest neighbour method (K-NN). In each situation under study, the average percentage of correct classification is considered. For all sample sizes, the study shows that EB method outperforms Classical and K-NN method in normal distribution whereas K-NN method outperforms EB and Classical method in heavy-tailed asymmetric distribution. In addition, the average percentages of correctly classified data in K-NN tend to be higher as the number of neighbourhoods (k) increases.