Abstract

Seismic patterns in volcanoes give an important information on volcanic activity and volcano monitoring demands accurate and robust systems to cover the volume of data registered by the monitoring sensors. This is why research on automatic volcanic pattern recognition techniques is ongoing. One of the most popular approaches is to design classifiers to discriminate patterns according to specific features, extracted from the signals. Literature presents different features and classification tools to perform these systems, reaching good results mainly in off-line experiments. This article presents the preliminary results of a proposal to design of a classifying structure where the features are selected, based on the expertise of the seismologists of the Observatorio Vulcanologico de los Andes Sur (OVDAS), the national net of volcano monitoring of Chile. The methodology implemented a multi-class classification structure based on Support Vector Machines (SVM), trained with the set of selected features. The classification structure reaches a mean error of 2.4% for all classes, achieving a better result than those obtained in previous studies where other feature selection criteria were adopted. This work verifies that the expert knowledge is relevant to select the features used in the design of the classification structures.