The characterization of regimes at an active volcano starts from a phase of data reduction, when spectral, dynamical and/or
stochastic parameters can be computed on successive time windows which duration determines a new time scale, that
typically goes from seconds to hours. The resulting parameter vectors (also called feature vectors) can then be used to try to
automatically classify the different phases of the volcanic activity, possibly also looking for precursors. This classification
can be done using many possible approaches, most of them using "machines" than have to be trained before they can be
applied to classify data. The training procedure can in turn be supervised or unsupervised. In this talk we present the
approach of Self Organizing Maps (SOM for short), an example of unsupervised machine, together with case studies of
application to volcanic tremor recorded at Raoul Island and Ruapehu volcanoes in New Zealand.
