Implementación de un método de agrupación de señales sísmicas generadas por el volcán Cotopaxi basado en aprendizaje automático no supervisado utilizando el modelo de mezcla gaussiana.

Abstract

The growing number of investigations in the machine learning area demonstrates its usefulness to apply itin variousfields ofstudy. For example, different types of volcanic earthquakes can be cluster to learn more about the dynamics of a volcano and its behavior. In this context, it would be possible to mitigate the risks that a possible volcanic eruption could cause, for the benefit of the people who live in the vicinity of a volcano. In the present work, clustering of volcanic seismic signals was carried out using unsupervised approaches such as the Gaussian Mixture Model. The clustering of signals was employed to obtain features. The seismic events that were used went through several steps before entering a supervised stage. Unsupervised approaches were found to allow learning a wide range of important features that are highly correlated with features provided by experts (baseline) over the years. Consequently, the results demonstrated that the unsupervised approaches that were compared (Gaussian Mixture Model and k-means) achieved statistically similar performances with respect to the baseline. Finally, the Gaussian Mixture Model outperformed the baseline features in noisy conditions.