Approach to solution of the task of attribute space formation with the time series prediction based on cluster analysis

The article is devoted to investigation of the problem of input attribute space formation for solution of the task of financial time series prediction with application of artificial neural networks. The new approach to solve this task is offered which is based on cluster analyses application. Artificial neural network of Kokhonen is applied as an instrument of cluster analyses. Comparison of the proposed approach with the well-known approaches to solution of this task is represented. Clustering quality coefficients and class recognition reliability are proposed. The efficiency of the offered approach with the financial time series prediction is experimentally proved, and namely, reduction of time for analyses and increase of reliability of the prediction obtained. Prospects for further development of this research trend are defined.