Institutional Repository of Vilnius University
Spatial contextual classification based on conditional distributions belonging to elliptical and exponential families /
Abstract
dc:descriptionWhen using supervised generative models in classification problems, the assumption of independence of observations is applied or the data is assumed to be modelled by a Gaussian Random Field (GRF). To solve classification problems for features with spatial contextual information with non-GRF, new classification methods are required. The research aims to construct supervised generative classification algorithms based on Bayes Discriminant Function (BDF) for features with spatial contextual information distributions belonging to exponential and elliptic families. Spatial contextual information is incorporated into the probability distribution of feature values and class labels. The properties of auto-models are used to construct exponential family models. BDF is defined by the ratio of logarithms of univariate conditional probability density (mass) functions. Expressions of error rates are obtained for the accuracy assessment. The proposed classifiers are investigated using simulated and real data, and compared using different class label functions. This work extends the application of supervised generative classification methods for features with distributions of spatial contextual information described by GRF and non-GRF models. An algorithm based on BDF has been developed, which allows using supervised generative classification models for feature values with distributions belonging to the exponential and elliptical families.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zikarienė, Eglė,
- Contributors dc:contributor
-
- Dučinskas, Kęstutis
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD142171369&prefLang=en_US