Catastrophe Prediction with Neural Network

摘要

The basis for this thesis is the definition of an optimal alarm system: a systemwhich, for a given probability of detecting a catastrophe, has the highest probability of correct alarm. It is shown that, theoretically, a neural network can be taught to approximate such an optimal alarm system arbitrarily well. The authors have made some comparative studies on simulated ARMA-processes, for which the optimal predictor can be derived theoretically. These studies confirm that a properly trained neural network can indeed approximate an optimal alarm system