Ieee Transactions on Signal Processing

| We examine a classiication problem in which seismic waveforms of Natural Earthquakes are to be distinguished from waveforms of Man-Made Explosions. We present an Integrated Classiication Machine (ICM) which is a hierarchy of Artiicial Neural Networks (ANN) that are trained to classify the seismic waveforms. In order to maximize the gain of combining the multiple ANNs, we suggest to construct a Redundant Classiication Environment which consists of severaìexperts' whose expertise depends on the diierent input representations they are exposed to. In the proposed scheme, the experts are ensembles of ANN, trained on diierent Bootstrap replicas 1]. We use various network architectures, diierent Time-Frequency decompo-sitions of the seismic waveforms and various smoothening levels in order to achieve a Redundant Classiication Environment. A conndence measure for the ensemble's classii-cation is deened based on the agreement (variance) within the ensembles 2] and an algorithm for a non-linear integration of the ensembles using this measure is presented. An implementation on a data set of 380 seismic events is described, where the proposed Integrated Classiication Machine (ICM) had classiied correctly 92% of the testing signals. The comparison we made with classical methods, indicates that combining a collection of ensembles of ANN's can be used to handle complex high dimensional classiica-tion problems.