Independent Component Analysis And Its Application In Financial Data Mining Based On Improved K-Means Cluster Algorithm

As an effective tool of data mining, Independent Component Analysis(ICA) has developed rapidly in recent years, which could reveal the influences and factors that hidden behind data. This article introduces the principles of classical models of ICA and combines with improved K-means cluster algorithm to analyze a stock and indicate the factor that affecting the closing price. The main work is as follows. Firstly, the article reviews different estimation methods of ICA model, then introduces the famous Fast ICA algorithm.Secondly, this paper gives concepts and principles of K-means cluster algorithm and demonstrates relative standards that evaluate the effect of the algorithm. Thirdly, the article applies improves K-means cluster algorithm to separate the outliers from normal data by using the contents of entropy, then evaluates the best method as final result through some criterions. Lastly, comparing the results of improved K-means cluster algorithm and the original one to examine the effect of the improved algorithm, then applying the Fast ICA algorithm based on negentropy to analyze the closing price of Ping An Bank and revealing the factor that influences the price behaviours.