A new kind of neural-based data computing approach for illness prediction has been presented in this paper. The purpose is to suggest a successful and efficient illness prediction method for the users to get substantial success in all kinds of massive data process, such as hospital illness cases. In this paper the benefits of combining two layered feed forward neural networks trained by back propagation on an identical data set are studied. Here, network diversity was achieved by the inherent randomness associated with the back-propagation algorithm's initialization of a network's weights. By case experiments of hospital illness, the technique has been tested effectively.