Improved Approach of Genetic Programming and Applications for Data Mining

Abstract

Genetic Programming (GP for short) is applied to a benchmark of the data fitting and forecasting problems. However, the increasing size of the trees may block the speed of problems reaching best solution and affect the fitness of best solutions. In this view, this paper adopts the dynamic maximum tree depth to constraining the complexity of programs, which can be useful to avoid the typical undesirable growth of program size. For more precise data fitting and forecasting, the arithmetic operator of ordinary differential equations has been made use of. To testify what and how they work, an example of service life data series about electron parts is taken. The results indicate the feasibility and availability of improved GP, which can be applied successfully for data fitting and forecasting problems to some extent.