Classification of Mount Etna (Italy) Volcanic Activity by Machine Learning Approaches

Abstract

Assessment of the ongoing activity of volcanoes is one of the key factors to reduce volcanic risks. In this paper, two Machine Learning (ML) approaches are presented to classify volcanic activity using multivariate geophysical data, namely the Decision Tree (DT) and K-Nearest Neighbours (KNN). The models were implemented using a data set recorded at Mount Etna (Italy), in the period 01 January 2011 – 31 December 2015, encompassing lava fountain events and intense Strombolian activity. Here a data set consisting of five geophysical features, namely the root-mean-square of seismic tremor (RMS) and its source depth, counts of clustered infrasonic events, radar RMS backscattering power and tilt derivative, was considered. Model performances were assessed by using a set of statistical indices commonly considered for classification approaches. Results show that between the investigated approaches the DT model is the most appropriate for classification of volcano activity and is suitable for early warning systems applications. Furthermore, the comparison with a different classifier approach, reported in literature, based on Bayesian Network (BN), is performed.