A Modified Method For Bayesian Prediction Of Future Order Statistics From Generalized Power Function

Loading...
Thumbnail Image
Date
2015-04
Authors
Omar, Almutairi Aned
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bayesian statistics is a statistical method that is widely used in many fields, including medicine, social and applied sciences. These fields occasionally have little or limited information about their populations. Therefore, using new techniques that require fewer samples while providing the same quality as the case of available samples is necessary. Bayesian prediction is a commonly used tool in Bayesian statistics. This study modifies three Bayesian prediction methods: one-, two- and multi-sample predictions. Bayesian prediction modified method does not require the many samples. Therefore, a future sample is a significant term in this thesis. Our Bayesian prediction modified method used a prediction for the future order statistics based on the observed ordered data, and predictive densities provided the Bayesian prediction intervals for the future order statistics. The standard generalized power function distribution serves as the basis for the three modified methods by applying Bayes' theory to achieve close lower and upper limits for the 95% and 99% Bayesian prediction intervals. The proposed intervals contributed to increasing the precision for the predictive value. The performance of the three modified methods is evaluated using the lower and upper limits from the observed sample and for the order statistic from the future sample. Two types of prior functions are used in Bayesian prediction: informative and non-informative priors, both of which use Bayes' theory. The numerical analysis illustrates lower and upper limits for the 95% and 99% Bayesian prediction intervals for the three modified methods, and the data set generated from the standard generalized power function distribution. Bayesian estimation is used to determine the shape, scale and location parameters. Bayesian estimators are suggested as the mean of the posterior distribution based on an informative or noninformative prior function. Both prior functions use a formula for the posterior distribution from Bayes theory to combine the likelihood function and prior function. The proposed Bayesian estimator is the mean of the posterior distribution based on the standard generalized power function distribution and a squared error loss function. In addition to this technique, a Bayesian criterion is used. The performance of the shape, scale and location estimators are evaluated with some types of prior distributions and used simultaneously with the Bayesian prediction, which, when compared, confirms the suitability and advantage of some types of prior distributions for estimation or prediction using the Bayesian method. The numerical analysis illustrates the proposed estimators derived from the data set generated from the standard generalized power function distribution.
Description
Keywords
Bayesian statistical decision theory
Citation