Entropy-based goodness-of-fit testsfor multivariate distributions


Dr. Öğr. Üyesi MEHMET SIDDIK ÇADIRCI

Tez Türü: Doktora

Tezin Yürütüldüğü Kurum: İngiltere

Tez Danışmanı: Nikolai Leonenko

Tezin Onay Tarihi: 2022

Tezin Dili: İngilizce

Desteklendiği Program: Diğer

Özet:

Entropy is one of the most basic and significant descriptors of a probability distribution.

It is still a commonly used measure of uncertainty and randomness in information

theory and mathematical statistics. We study statistical inference for Shannon

and Rényi’s entropy-related functionals of multivariate Gaussian and Student-t distributions.

This thesis investigates three themes. First, we provide a non-parametric

test of goodness-of-fit for a class of multivariate generalized Gaussian distributions

based on maximum entropy principle via using the k-th nearest neighbour (NN) distance

estimator of the Shannon entropy. Its asymptotic unbiasedness and consistency

are demonstrated. Second, we show a class of estimators of the Rényi entropy based

on an independent identical distribution sample drawn from an unknown distribution

f on Rm. We focus on the maximum Rényi entropy principle for multivariate

Student-t and Pearson type II distributions. We also consider the entropy-based test

for multivariate Student-t distribution using the k-th NN distances estimator of entropy

and employ the properties of entropy estimators derived from NN distances.

Third, we introduce a new classes of unimodal rotational invariant directional distributions,

which generalize von Mises-Fisher distribution. We propose three types of

distributions in which one of them represents the axial data. We provide all of the

formula together with a short computational study of parameter estimators for each

new type via the method of moments and method of maximum likelihood. We also

offer the goodness-of-fit test to detect that the sample entries follow one of the introduced

generalized von Mises-Fisher distribution based on the maximum entropy

principle using the k-th NN distances estimator of Shannon entropy and to prove its

L2-consistence.