Building Machine Learning Models for Long-Period and Volcano-Tectonic Event Classification

The proper identification of several types of volcanic seismic events can be related to the intrinsic behavior of a volcano, and it could be useful to provide an early alarm in the case of an impending eruption. Among different recorded seismic events, the long-period and volcano-tectonic stand as the most important events to track since their occurrence increase may help to forecast possible eruptions. Thus, the correct classification of both types of seismic events could increase the safety of peoples living around the volcano. In this sense, this paper proposed a comprehensive algorithm for building classification models to classify long-period and volcano-tectonic seismic events based on a suitable combination of five well-known machine learning classifiers that provide the highest performance over the area under receiver operating characteristic curve. The method explores different machine learning model combinations to find the most adequate in the context of long-period and volcano-tectonic seismic event classification. The proposed method was validated on an experimental dataset, containing 587 long-period and 81 volcano-tectonic seismic events recorded from the Cotopaxi Volcano in Ecuador. The random forest classifier with 100 tree based predictors demonstrated to be the best classification model. According to the Wilcoxon Statistical test, the proposed method was effective in providing competitive models for the classification of volcano seismic events.