Auto-associative multi-layered neural networks for the classification of seismic signals

Autoassociative multilayered neural networks were used to validate the seismic shock classification performed by expert seismologists on the basis of somewhat heuristic criteria. It was possible to evaluate the degree to which each event in the training data set belonged to each class, independently of the seismologist's suggestions, avoiding errors when heuristic criteria were not adequate. The networks allow recognition of events incorrectly classified, so that they can be reanalyzed by the human expert.<<ETX>>