Experimental design, expert system and neural network approaches: comparison for the choice of parameters

The choice of the parameters in a system can be supported by several methods, which depend on the type of the available knowledge. In the experimental case of the parametrization of fuzzy dispatching rules in a scheduling software, three methods are tested: experimental design, fuzzy expert system and neural network. In an experimental design, combinations of different values of the parameters are tested in a way that allows to distinguish the respective influence of each parameter. If a deep knowledge on the influence of the parameters on the outputs of the system is available, and if this knowledge can be expressed through production rules, an expert system may be designed. If these conditions are not verified, but if examples of correct set-ups in various cases can be found, a neural network can give a satisfying answer to the problem. A possible implementation of each method is shortly described, and their fields of interest discussed