ICA-Based Potential Significant Feature Extraction for Market Forecast

• Published in: 2006 International Conference on Computational Inteligence for Modelling Control and Automation and International Conference on Intelligent Agents Web Technologies and International Commerce (CIMCA'06) p. 176
• Main Authors: Ling Huang, Jinhong Zhong
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2006
• Subjects: Costs; Demand forecasting; Economic forecasting; Feature extraction; Finance; Independent component analysis; Independent Component Analysis (ICA); Neural networks; prediction of dynamic data; Principal component analysis; Principal Component Analysis (PCA); Radial Basis Function Neural Network (RBF NN); Risk management; Signal resolution
• ISBN: 0769527310; 9780769527314
• DOI: 10.1109/CIMCA.2006.116

Most of market demands are greatly dynamically changed such as in finance, industries and etc. In order to adapt to the highly dynamic changes of market and reduce cost and risk, we need to develop effective management strategies based on accurate market demand forecast. Many of the traditional forecasting methods, such as the neural network, ARIMA, which based on historical data, cannot find out the hidden remarkable features behind the data, thereby affecting the accuracy of forecasting. In this paper, ICA as one of the most popular signal decomposition technologies in recent years is introduced to excavate the potential information of market for better prediction of dynamic changes in market demand and mining of margin customer, where ICA plays an important role of preprocessing. The experiments show that the prediction based on ICA preprocessing is superior to direct prediction by neural network, and successful to excavate potential customers for better market directing. In addition, we also address the essential difference of dimension reduction using PCA and ICA show that these two approaches are different at the aspect of sensitivity to dimensions although they both are pre-processing methods of dynamic data, even if the accumulative contribution rate of ICA is 7.4% less than that of PCA the former still attains the same prediction results as the latter.