Nonlinear modeling for time series based on the genetic programming and its applications

This paper deals with clustering of segments of stock prices by using nonlinear modeling system for time series based on the Genetic Programming (GP). We apply the GP procedure in learning phase of the system where we improve the nonlinear functional forms to approximate the models used to generate time series. The variation of the individuals with relatively high capability in the pool can cope with clustering for various kinds of time series which belong to the same cluster similar to the classifier systems. As an application, we show clustering of artificially generated time series obtained by expanding or shrinking by transformation functions. Then, we apply the system to clustering of 8 kinds of segments of real stock prices.