Time series forecasting by hybrid artificial intelligence architecture and its application

This paper proposed a hybrid model to improve the single SVR model. The hybrid model that is composed of neural network and support vector machine (SVR) has a two-stage neural network architecture. In the first stage, self-organizing feature map (SOM) can be used as a clustering algorithm to partition the whole input space into several disjointed regions. In the second stage, based on the principle of least error, SVR which best fit partitioned regions are constructed by finding the most appropriate kernel function. The application of GDP, CPI and Total Foreign Trade Volume prediction shows that SOM-SVR models achieve significant improvements in the generalization performance compared with the single SVR model. Additionally, the SOM-SVR models also converge faster.