ICA Adaptive Algorithm And Its Application In The Financial Data Mining

Independent component analysis (ICA) is a powerful tool for data processing, which developed in recent years. It has been widely used in various fields. This paper introduces the development process of ICA briefly and discusses the objective functions and algorithms of ICA in detail. Two variable step-size programs are combined to improve the efficiency of the adaptive algorithm of ICA. Finally, ICA is applied to financial data mining combined with clustering. The main work of this paper can be summarized as follows:1. Proposing the ICA maximum likelihood estimation adaptive algorithm based on line search and equivariant program. Line search which is the optimization of single variable function is one of the methods for solving the problem of unconstrained nonlinear programming. The equivariant program is the method which can speed up convergence speed and reduce the misadjustment error in the steady state simultaneously by building a nonlinear function relationship between step size and the difference among separating matrixes. The ICA algorithms of fixed step have natural defect that the step size can only guarantee one of the convergence and stability to people's expectatioa While this paper improves the ICA maximum likelihood estimation algorithm of fixed step by using line search and equivariant program, which overcomes the deficiency of original algorithm in the steady stage and sudden imitative environment and reaches the adaptive effect. The simulation experiment results show that the two modified algorithms are feasible and effective.2. Proposing the method of time series clustering based on ICA. Clustering is one of the unsupervised learning methods. Its goal is to identify structure in an unlabeled data set by objectively organizing data into homogeneous groups where the within-group-object similarity is minimized and the between-group-object dissimilarity is maximized. Firstly, this paper optimizes of the selection of the initial cluster centers through the hierarchical algorithm to give a modified k-means algorithm. Then making it combine with FastICA algorithm and the method of time series clustering based on ICA is obtained. Finally, This paper applies the above method to financial data mining. 40 stocks are clustered and deep reasons hidden behind the stock market are analyzed. The simulation experiment results show that the method can resolve the redundant problem of financial time series and response some important features of financial market.