Volatilite aktivasyon fonksiyonu ile zaman serisi tahmini
Küçük Resim Yok
Tarih
2022
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
İstanbul Beykent Üniversitesi
Erişim Hakkı
info:eu-repo/semantics/openAccess
Özet
Zaman serisi analizi zor bir o kadarda kritik bir konudur. Zaman serisi analizi toplumun birçok kesimi tarafından kullanılmakta kritik kararlarda önemli bir rol oynamaktadır. Bununla birlikte zaman serisi analizi ile ilgili modern çalıĢmalar çok kısıtlı ve yeterli değildir. ÇalıĢma kapsamında zaman serisi analizi için yapay zeka tekniklerinden yararlanılmıĢtır. Yapay zeka teknikleri kullanılırken hiper parametrelerin belirlenmesi, verilere göre farklı aktivasyon fonksiyonlarının kullanılması ihtiyacı, aktivasyon fonksiyonlarındaki problemler, modelin farklı noktalarında aktivasyon fonksiyonu kullanılması, aktivasyon fonksiyonun veri türüne yakınsamaması gibi problemler bulunmaktadır. ÇalıĢma kapsamında bu problemlerin çözümü ve zaman serisi analizi üzerine önerilerde bulunulmuĢ bu öneriler ispatlanmıĢtır. Problemlerin çözümü için modifiye aktivasyon fonksiyonları ve kollektif sınıflandırma ile yapay zeka algoritmalarının yarıĢtırılması yöntemleri kullanılmıĢtır. Leaky ReLU ve Sigmoid fonksiyonu üzerine modifiye iĢlemi yapılmak suretiyle Modifiye Sigmoid ve Modifiye Leaky ReLU fonksiyonları önerilmiĢtir. Modifiye sigmoid fonksiyonu ile vanishing gradyan ve ReLU fonksiyonunda bulunan linear olma problemleri çözülmüĢtür. Literatürde önerilen finansal enstrüman tahmin modellerini kapsayıcı bir mimari oluĢturularak bu kapsamda bir uygulama geliĢtirilmiĢtir. Önerilen yöntemeler literatür çalıĢmaları ile kıyaslanarak yöntemlerin doğrulukları ve uygulanabilirlikleri ispatlanmıĢtır. Yapılan çalıĢma sonucunda literatüre yeni aktivasyon fonksiyonları ve yapay zeka algoritmaları katkılarında bulunulmakla birlikte problemlerin çözümüne alternatif yöntemler sunulmuĢtur. Bu çalıĢmalar sonucu göstermiĢtir ki önerilen yöntemler uygulanabilir ve tutarlıdır.
Time series analysis is a difficult and critical issue. Time series analysis is used by many parts of the society and plays an important role in critical decisions. However, modern studies on time series analysis are very limited and not sufficient. Within the scope of the study, artificial intelligence techniques were used for time series analysis. When using artificial intelligence techniques, there are problems such as the determination of hyperparameters, the need to use different activation functions according to the data, the problems in the activation functions, the use of the activation function at different points in the model, the activation function not converging to the data type. Within the scope of the study, suggestions were made on the solution of these problems and time series analysis and these suggestions were proven. For the solution of the problems, modified activation functions and collective classification and artificial intelligence algorithms competing methods are used. Modified Sigmoid and Modified Leaky ReLU functions have been proposed by modifying the Leaky ReLU and Sigmoid functions. With the modified sigmoid function, the vanishing gradient and linearity problems in the ReLU function are solved. An application has been developed in this context by creating an architecture that covers the financial instrument forecasting models suggested in the literature. The accuracy and applicability of the methods have been proven by comparing the proposed methods with the literature studies. As a result of the study, new activation functions and artificial intelligence algorithms were contributed to the literature, and alternative methods were presented for the solution of the problems. The results of these studies have shown that the proposed methods are applicable and consistent.
Time series analysis is a difficult and critical issue. Time series analysis is used by many parts of the society and plays an important role in critical decisions. However, modern studies on time series analysis are very limited and not sufficient. Within the scope of the study, artificial intelligence techniques were used for time series analysis. When using artificial intelligence techniques, there are problems such as the determination of hyperparameters, the need to use different activation functions according to the data, the problems in the activation functions, the use of the activation function at different points in the model, the activation function not converging to the data type. Within the scope of the study, suggestions were made on the solution of these problems and time series analysis and these suggestions were proven. For the solution of the problems, modified activation functions and collective classification and artificial intelligence algorithms competing methods are used. Modified Sigmoid and Modified Leaky ReLU functions have been proposed by modifying the Leaky ReLU and Sigmoid functions. With the modified sigmoid function, the vanishing gradient and linearity problems in the ReLU function are solved. An application has been developed in this context by creating an architecture that covers the financial instrument forecasting models suggested in the literature. The accuracy and applicability of the methods have been proven by comparing the proposed methods with the literature studies. As a result of the study, new activation functions and artificial intelligence algorithms were contributed to the literature, and alternative methods were presented for the solution of the problems. The results of these studies have shown that the proposed methods are applicable and consistent.
Açıklama
Anahtar Kelimeler
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrol, Computer Engineering and Computer Science and Control, Bilim ve Teknoloji
