Search

This paper considers the realization of an intelligent digital signal processing system for representing time series and its application to the classification of stock prices. The system utilizes an effective mechanism to detect the nonstationary part (transient waves) which includes important information on the time series based on the Gabor representation and the knowledge representation of waveforms using an expert system. In the subsystem for digital signal processing, an adaptive ARMA model is fitted to the time series obtained by subtracting the moving average from the original time series to generate the time series containing enhanced transient waves. The Gabor representation is applied to classify the kinds of transient waves. A set of codebooks for the spectrum of transient waves is used to identify the kind of transient wave. In the subsystem of the expert system, the characteristics of the time series obtained by digital signal processing and other features recognized by conventional methods such as trend lines are transferred to the frames so that the recognition rules of experts are applied to classify the time series. As an applications, 194 time series of stock prices are recognized and classified into 11 categories. [ABSTRACT FROM AUTHOR]