SVR for time series prediction
Support Vector Machines: Data Analysis, Machine Learning and Applications, NOVA Publications , ss.117-130, 2011
- Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
- Basım Tarihi: 2011
- Yayınevi: NOVA Publications
- Sayfa Sayıları: ss.117-130
- Anahtar Kelimeler: Artificial neural networks (ANN), Feedback, Financial time series prediction (FTSP), Global, Istanbul stock exchange (ISE), Smoothed piecewise neural models, Support vector predictors (SVP), Volatility
- Boğaziçi Üniversitesi Adresli: Evet
Özet
This chapter makes a comparison of support vector machines and neural prediction models for the financial time series prediction problem. Support Vector Machines (SVMs) are very successful at classification and regression tasks and provide a good alternative to neural network architectures. Unlike neural networks training which requires nonlinear optimization with the danger of getting stuck at local minima, the SVM solution is always unique and globally optimal. By considering this unique and globally optimal property, we have used SVM as a regression approach to financial time series forecasting. This chapter discusses the advantages and disadvantages of each model by using a real-world data: 22 years Istanbul Stock Exchange ISE 100 index data from 1988 to 2010. Several performance metrics are used to compare these models including regression metrics, prediction trend accuracy metrics and special metrics such as Spearman's and Kendall's rank correlation coefficients Finally, it is observed that the support vector predictors becomes advantageous in capturing volatility in index return series when it is compared to global and feedback neural models. © 2011 by Nova Science Publishers, Inc. All rights reserved.