Determination of inspection order for classifying new samples by neural networks

The aim of this paper is to propose an idea for determining the inspection order for classifying new samples by neural-network-based classification systems. In real world classification problems such as medical diagnoses, inspection costs for measuring many inspection items can not be negligible. Therefore it is desirable to classify new samples by measuring a small number of inspection items. In this paper, first we propose a method for classifying new samples by partial information on input values in neural-network-based classification systems. The proposed method is based on the interval representation of unknown (i.e., unmeasured) input values. Next we propose an idea for determining the inspection order of input values for new samples. Last we illustrate the proposed approach by computer simulations on the iris data