یک رگرسیون بردار پشتیبانی ICA برای پیش بینی قیمت های نفت خام  An ICA-based support vector regression scheme for forecasting crude oil prices

1. Introduction Crude oil is one of the most actively traded commodities in the world (Alvarez-Ramirez et al., 2012). The large fluctuations of crude oil prices affect the economic growth of importing and exporting countries as well as regional security and stability (Wu and Zhang, 2014). Recent decades have seen the more frequent fluctuation of crude oil prices, which attracted the concern from both market participants and governmental regulators (Zhang, 2013). Undoubtedly, accurate oil price forecasting is of strategic significance in multiple aspects such as determining the timing for crude oil importing and ensuring economic security (Zhang and Wang, 2013). The intrinsic complex features of oil prices and the uncertainty in economic policy pose big challenge on the accurate forecasting of crude oil prices (Bekirosa et al., 2015). Many scholars have thus contributed to develop novel methods and models for improving the accuracy of crude oil price forecasting. Fan and Li (2015) provided a relatively comprehensive review of major crude oil price forecasting models and found that artificial intelligence models (e.g. neural networks and support vector machines) had received increased attention. Recent methodological developments of artificial intelligence-based forecasting models can be found in Zhu and Wei (2013), Yu et al. (2014), Azadeh et al. (2012; 2015), Barunik and Krehlik (2016), Chen and Chen (2016), Mostafa and El-Masry (2016), and Oztekin et al. (2016). In crude oil price forecasting, Jammazi and Aloui (2012) combined wavelet decomposition and artificial neural network to achieve better forecasting performance. He et al. (2012) showed the effectiveness of a wavelet decomposed ensemble model. Tang and Zhang (2012) developed a multiple wavelet recurrent neural network simulation model to analyze crude oil prices. Guo et al. (2012) proposed an improved support vector machine (SVM) model by using genetic algorithm to optimize the parameters. Zhang et al. (2015) proposed a hybrid method by combining SVM with ensemble empirical mode decomposition and particle swarm optimization models to improve the forecasting performance. Wang et al. (2016) recently proposed a Markov switching multifractal volatility model to forecast crude oil return volatility.