AUTOMATIC RANKING OF FUZZY NUMBERS WITH THE CRITERION OF A DECISION-MAKER LEARNT BY AN ARTIFICIAL NEURAL-NETWORK

In a previous work, we indicated that Artificial Neural Networks (ANN) would be able to learn to compare fuzzy numbers as a real decision maker does. In this paper, we describe in detail the experiment that we have developed to that goal, and in which we have obtained good results. We apply this trained ANN to some decision problems with fuzzy environment, by means of the automatic ranking of the decision problem utilities, performed as trapezoidal fuzzy numbers. So, we use the trained ANN as a personal method to compare fuzzy numbers. We have trained a multilayer feedforward ANN with the criterions (to compare fuzzy numbers) of three people, each with different characteristic, using the backpropagation algorithm and different structures. Then we use this trained ANN to rank a set of fuzzy numbers which can be considered as utilities of decision problems with fuzzy environment, hence enabling us to make the best choice. Several examples are shown also.