The Research Of Feature Extraction Based On Independent Component Analysis

Machine Learning aims to tackle large and complex problem domains, the problem of drawing attention to the most relevant information in potentially vast amounts of data and extracting the essential features from the various things have become increasingly more relevant and important. For a long time, the focus of feature extraction is on how to use the smaller data to represent the differences between the different things, in other words, the problem of feature extraction to a certain extent is the technique of reducing data dimension.Performance and the cost of classification are sensitive to the choice of the features used to construct the classifier.As a result, feature extraction plays an important role in classification tasks. The main goal of all feature extraction algorithms is to reduce the computational complexity and improve classification performance by discarding less relevant or redundant features.The main work of this thesis is to introduce the Independent Component Analysis as a method of feature extraction. The principle of ICA algorithm is to find the potential mutual independent components, to remove higher-order redundance between components and to extract the independent original signals source by analyzing the high-order statistical correlation of the multidimensional data. In the thesis, we introduce the basic principles and related algorithms from the point of the statistical methods and information theory, and compare with the other familiar methods of feature extraction.In the thesis, we use the FastICA algorithm to extract the features of some datasets of Machine Learning Repository in different condition, at the end of this thesis, we utilize the Support Vector Machine and the other classifier to classify the features which have been extracted. By comparing the accurancy of the classification, we can prove the efficiency of the ICA in feature extraction.