Publication: Sürekli Bayesci Ağlar için Genelleştirilmiş Lineer Modeller Yaklaşımı
Abstract
Bayesci Ağlar, değişkenler arasında olası ilişkilerin belirlenmesi ve karar verme sürecinde hem uzman hem de verilere dayalı grafiksel bir yöntemdir. Bu yöntem kesikli veriler üzerinde oldukça yaygın kullanılmaktadır. Sürekli verilerin mevcut olması durumunda, Klasik Bayesci Ağlar yönteminin kullanımı ile oluşturulan modeller doğru ve güçlü performans gösteremezler. Ancak, gerçek hayatta ekonomi, sağlık, biyoloji, kimya vb. gibi birçok alandaki problemlerin içerdiği veri setleri sürekli değişkenlerden oluşmaktadır. Genelleştirilmiş Lineer Modeller yaklaşımı ile oluşturulmuş Bayesci Ağlar, birçok problemin çözümünü, karar verme ve planlama sürecini daha ekonomik, hızlı, etkin, ve güvenilir bir biçimde gerçekleştirilebilmektedir. Bu amaçla kullanılan Tweedie sınıfına ait olan Tweedie Bayesci Ağlar yöntemi, yakın zamanda keşfedilmiş ve literatüre kazandırılmış bir yöntemdir. Bu çalışmada, yine Genelleştirilmiş Lineer Modeller çatısı altında Bayesci Ağlar incelenmiş ve Normal Bayesci Ağlar'a karşı olan gücü ve üstünlüğü hem gerçek bir veri üzerinde hem de simülasyon çalışmalarıyla sunulmuştur. Ayrıca, konum, ölçek ve şekil için Genelleştirilmiş Toplamsal Model yöntemi ile Bayesci Ağlar entegre edilerek, sürekli verilerle ile oluşturulabilecek alternatif yöntem test edilmiştir. Çalışma kapsamında kullanılan veri seti ile oluşturulan modeller karşılaştırılarak yöntemler sunulmuştur.
Bayesian Networks are both expert and data-based graphical methods in the determination of possible relationships between variables and in the decision-making process. This method is widely used on discrete data. In the presence of continuous data, models created using the Classical Bayesian Networks method cannot perform accurately and strongly. However, in real life it can be used in economics, health, biology, chemistry, etc. The data sets included in the problem in many areas such as these consist of continuous variables. Bayesian Networks, created with the Generalized Linear Models approach, can perform the solution of many problems, decision making and planning processes in a more economical, fast, effective, and reliable way. The Tweedie Bayesian Networks method, which belongs to the Tweedie class used for this purpose, is a method that was recently discovered and brought to the literature. In this study, Bayesian Networks are examined under the umbrella of Generalized Linear Models and their strength and superiority over Normal Bayesian Networks are presented both on a real data and simulation studies. In addition, the alternative method that can be created with continuous data by integrating the Generalized Additive Model method and Bayesian Networks for position, scale and shape is tested. The methods are presented by comparing the models created with the data set used in the study.
Bayesian Networks are both expert and data-based graphical methods in the determination of possible relationships between variables and in the decision-making process. This method is widely used on discrete data. In the presence of continuous data, models created using the Classical Bayesian Networks method cannot perform accurately and strongly. However, in real life it can be used in economics, health, biology, chemistry, etc. The data sets included in the problem in many areas such as these consist of continuous variables. Bayesian Networks, created with the Generalized Linear Models approach, can perform the solution of many problems, decision making and planning processes in a more economical, fast, effective, and reliable way. The Tweedie Bayesian Networks method, which belongs to the Tweedie class used for this purpose, is a method that was recently discovered and brought to the literature. In this study, Bayesian Networks are examined under the umbrella of Generalized Linear Models and their strength and superiority over Normal Bayesian Networks are presented both on a real data and simulation studies. In addition, the alternative method that can be created with continuous data by integrating the Generalized Additive Model method and Bayesian Networks for position, scale and shape is tested. The methods are presented by comparing the models created with the data set used in the study.
Description
Keywords
İstatistik, Bayes İstatistiksel Karar Teorisi, Gama Dağılımı, Gauss Dağılımı, Normal Dağılım, Normal Olmayan Dağılım, Statistics, Bayesian Statistical Decision Theory, Sürekli Zaman Genelleştirilmiş Öngörülü Denetim, Gamma Distribution, Gaussian Distribution, Veri Yorumlama-İstatistiksel, Normal Distribution, Nonnormality Distribution, Önsel Dağılım, Continuous-time Generalized Predictive Control, Data Interpretation-Statistical, İstatistik Dağılımlar, Prior Distribution, Statistical Distributions, İstatistiksel Dağılım, Statistical Distribution
Citation
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
148
