An experimental evaluation of different methods for handling missing inputs.

This paper presents mt empirical comparison among four different methods of handling missing inputs. Eight different classification problems me used for the comparison, and the performance of the methods is obtained for several percentages, 0%, 5%, 10%, 30% and 40% of missing inputs in the problem. There is one method which has the best performance in the eight different problems. The method is based on generalization of Backpropagation to interval arithmetic and a codification during training of every missing input by the interval [0, 1].