An agent-based approach to identification of prediction models

This paper presents an agent-based approach to identification of prediction models in two-dimensional data spaces. A number of agents are sent to the two-dimensional data space that people want to investigate. At the micro-level, every agent tries to build a local linear model by competing with others, and then at the macro-level all surviving agents build the global model by cooperating with each other. And a genetic algorithm is introduced for improving the global model built by the agents. Two examples that apply this approach are given. The advantages of this approach are it does not need people to give a certain formula in advance; and most of time, it can give more precise prediction models than those given by traditional methods.