Comparing evaluation methods based on neural networks for a virtual reality simulator for medical training

Several approaches for on-line training evaluation in virtual reality simulators have been proposed for performance classification of a trainee in pre-defined classes of training. However, how to choose the best method to evaluate a particular kind of training remains as an unsolved problem. Some evaluation methods are based on neural networks. Particularly, two of them use backpropagation trained multilayer perceptron neural network and evolving fuzzy neural networks. The present paper provides a comparison between these methods using a real case of training simulator for bone marrow harvest. Results obtained are analysed using classification matrices, graphical of classifications mistakes and the Kappa Coefficient. Based on the performance observed, some considerations about the choice between these two methods are provided.