Search

Basit ve çoklu doğrusal regresyon çözümlemesinde klasik metotlardan yararlanılıyorsa, modele ilişkin parametrelerin tahmini bazı varsayımlara dayanır. Doğrusal regresyon modellerinin bilinmeyen parametrelerinin tahmininde verilerin normal dağılım dışında bir dağılıma sahip olması durumu tahmin sürecinde klasik çözümlemelerin dışına çıkılmasını gerektirir. Böyle durumlarda bulanık mantığa dayalı çözümleme yöntemleri alternatif yöntemler olarak kendini göstermektedir. Bu çalışmada, bağımsız değişkenlerden herhangi birinin Pareto dağılımına sahip olması ve veri setinde aykırı gözlemlerin mevcut olması durumunda, kurulacak çoklu doğrusal regresyon modelinin bilinmeyen parametrelerini tahmin etmek için bir algoritma önerilerek, algoritmadan elde edilen tahminler literatürde yer alan mevcut yöntemlerden elde edilen tahminler ile karşılaştırılmıştır. / if classical methods are being employed in simple and multiple linear regression analysis, estimation of parameters related to the model is based on some assumptions. If the data have any distribution other than normal distribution in estimation of unknown parameters of linear regression models, the analysis has to go beyond classical analysis. In such cases analysis methods based on fuzzy logic manifest themselves as alternative ways. In this paper, an algorithm has been proposed in order to be able to estimate the unknown parameters of multiple linear regression model in the case that any independent variable has a Pareto distribution and incompatible observations exist in data set and estimations obtained from this algorithm have been compared with estimations obtained from the methods existing in the literature.

In this study, the biased estimators for the shape parameter of classical Pareto distribution by using the maximum likelihood estimator and the unbiased estimator are proposed. Mean square errors of these biased estimators are smaller than mean square errors of maximum likelihood estimator and the unbiased estimator. Then the unbiased estimator and the maximum likelihood estimator are compared with obtained biased estimators by means of simulation study and conclusions are interpreted. [ABSTRACT FROM AUTHOR]

This paper presents general information on simulation with Pareto distribution and introduces an algorithm to generate random numbers by simulation method. A computer code has been developed for this algorithm by using Delphi programming language. Ten thousand random numbers for Pareto distribution have been generated using this program, and the chi-square goodness-of-fit at five percent significance level has been performed. [ABSTRACT FROM AUTHOR]