Automatic classification of volcanic tremor using Support Vector Machine

Abstract

A system for automatic recognition of different volcanic activity regimes based on supervised classification of volcanic tremor is proposed. Spectrograms are calculated from volcanic tremor time-series, separated into four classes, each assumed as representative of a different state of volcanic activity, i.e., pre-eruptive, eruptive, lava fountains, and post-eruptive. As classification features, the spectral profiles obtained by averaging each spectrogram along its rows are chosen. As supervised classification strategy, the Support Vector Machine (SVM) classifier is adopted. Evaluation of the system performance is carried out on volcanic tremor data recorded at Mt Etna during the eruptive episodes of July-August 2001. The leave-one-out classification accuracy achieved is of about 94%.