Seismic events discrimination by neuro-fuzzy-based data merging

Abstract

This article involves an original method to classify low magnitude seismic events recorded in France by a network of seismometers. This method is based on the merging of high-level data with possibly incomplete low-level data extracted from seismic signals. The merging is performed by a multi-layer neural network. A fuzzy coding is applied to the neural network's inputs to process efficiently incomplete data. The results reveal that the fuzzy coding coupled with the data merging increases the correct classification rate to more than 90% even when the database contains missing values.