Independent Component Analysis And Its Applications In Data Mining

Recently, with the Blind Signal Separation problem the Independent Comp-onent Analysis (ICA) has been highlighted in statistic signal proposing, and become the more powerful high-dimensional multivariate data analysis tool. Thebasic principal of ICA is to find the independent hidden information through analyzing the high-order statistic relation of observed data, and achieve the goalof getting rid of the redundant high order component and extracting the inde-pendent source data. Compared with other methods, ICA has the characteristics by which we can recover the independent hidden source from observed data i-n the condition that there is no information about the source data and the mi-xing mechanism. The special advantage makes ICA more and more widely ap-plied in image feature extraction, compression, pattern recognition and data mi-ning etc. Data Ming is the hotspot in recent computer research, that is an unordina-ry pattern recognition process. The goal of data mining is to find the correct,novel, valuable and interpretable pattern from the mass data. The mass scientif-ic data not only has the great quantity, but has the high dimensional complica-ted features. These characteristics challenge t-he traditional data mining technol-ogy. So we need the new feature processing technology, but these knowledge can hardly be found in normal data mining book. This paper will introduce the optimizing technology of features in detail, especially the ICA feature reducing technology. The main work of this paper focus on the following aspects:1. Introduce the basic idea of optimizing technology of feature, and the two kinds of optimizing criterion. Give the decision tree al- gorithm of feature selection.2. Introduce the theory of ICA in detail, including the problem and the future of ICA.3. Introduce the statistic theory and information theory related to I- CA.4. Research the optimizing algorithms of ICA, discuss the two kind s representative algorithms based on nongaussian maximum or I- nformation entropy maximum, finish the realization work of Fast-...