DEALING WITH MISSING VALUES IN NEURAL-NETWORK-BASED DIAGNOSTIC SYSTEMS

Backpropagation neural networks have been applied ;to prediction and c lassification problems in many real world situations. However, a drawb ack of this type of neural network is that it requires a full set of i nput data, and real world data is seldom complete. We have investigate d two ways of dealing with incomplete data - network reduction using m ultiple neural network classifiers, and value substitution using estim ated values from predictor networks and compared their performance wit h an induction method. On a thyroid disease database collected in a cl inical situation, we found that the network reduction method was super ior. We conclude that network reduction can be a useful method for dea ling with missing values in diagnostic systems based on backpropagatio n neural networks.