Time Series Forecasting by ANN and SVM Hierarchical Architecture
Time series forecasting is a very important task in machine learning field. Because time series are inherently noise and non-stationary, it is difficult to get an accurate forecasting result. In this paper, a novel hierarchical learning architecture is proposed according to divide-and-conquer principle, the forecasting accuracy is improved. This novel hierarchical architecture is formed by Kohonen neural network and SVM (Support Vector Machine) as basic building blocks. The Kohonen network is used as a classifier, which partition the whole input space into several distinct feature regions. Then, the SVM can best fit disjoint regions by using the most appropriate kernel function. The sunspot data and financial time series are evaluated in the experiment. The result shows that the hierarchical architecture achieves higher accuracy in comparison with single SVMs and ANN models.
time series artificial neural network support vector machine, forecasting, hierarchical architecture.
WU Wei LIU Hongbin WANG Xuan LV Jiake
Department of Computer and Information Science University of Southwest Chongqing, 400715, China Department of Resource and Environment University of Southwest Chongqing, 400715, China
国际会议
武汉
英文
25-28
2007-07-25（万方平台首次上网日期，不代表论文的发表时间）