A backpropagation neural network for risk assessment

The authors investigate the credibility of the neural network approach as a viable tool in the field of developmental toxicity risk assessment. A three-layer artificial neural network (ANN) was trained using backpropagation. The topology of the network was decided based on a set of trials and errors. This network was trained to perform risk assessment on a set of toxicological data and give a decision like the decision given by experts. The assessment ability of the resulting network was compared with the statistical approach of discriminant analysis and the superiority of the neural network approach was established