Finding patient cluster attributes using auto-associative ANN modeling

Auto-associative artificial neural network models can be trained from medical databases using efficient training algorithms. Clustering can be achieved by grouping examples with a similar pattern in the hidden layer neurons' outputs. An example of the clustering of a large set of New Zealand asthma symptoms questionnaire data is presented. The results show that good clustering is feasible and new knowledge can be inferred from the means of the examples' attributes included in each cluster.