Knowledge discovery for geographical cellular automata

This paper proposes a new method for geographical simulation by applying data mining techniques to cellular automata. CA has strong capabilities in simulating complex systems. The core of CA is how to define transition rules. There are no good methods for defining these transition rules. They are usually defined by using heuristic methods and thus subject to uncertainties. Mathematical equations are used to represent transition rules implicitly and have limitations in capturing complex relationships. This paper demonstrates that the explicit transition rules of CA can be automatically reconstructed through the rule induction procedure of data mining. The proposed method can reduce the influences of individual knowledge and preferences in defining transition rules and generate more reliable simulation results. It can efficiently discover knowledge from a vast volume of spatial data.