The Research Of Exchange Rate Forecasting Based Compressed Sensing And Weighted Rough Set Clustering

In this paper,we start with the two learning methods of machine learning:regression of supervised learning and clustering of unsupervised learning.The difference between two methods is facing different types of data.In regression,due to the complexity and uncertainty of fluctuations in exchange,more accurate methods should be studied.In this paper,we first apply compressed sensing based support vector regression learning for exchange rate forecasting.Compressed sensing denoising is an procedure that denois-ing original data for preprocessor,and support vector regression is employed to conduct prediction for the cleaned data.For experimental study,we select data for RMB against US dollar from the People ' s Bank of China as sample data.Comparing with other de-noising tools such as wave denoising,compressed sensing based support vector regression learning is a better choice for exchange rate forecasting.In clustering,original rough set attribute reduction is too effective to handle with the data which has few attribute.The over-reduction may cause information loss espe-cially facing the data which has few attribute.Besides,neighborhood rough set is more compatible when dealing with continuous variable.So,we apply neighborhood rough set instead of rough set to attribute reduction when facing continuous variable.As for the ignorance of attribute's importance degree,we demonstrate a improvement method which combined neighborhood rough set and standard deviation of attributes value,weighted the clustering distance.Then we use UCI database for experimental study,which proved the new method's effectiveness.