A Machine Learning Approach for Classifying Micro-Earthquakes at Llaima Volcano

Automated systems play a key role in the development of early warning mechanisms with the objective of preserving lives and securing regions susceptible to volcanic activity. The aim of this article is to develop intelligent algorithms based on Machine Learning for the multiclass classification of micro-earthquakes originated at Llaima volcano, including tectonic earthquakes, long-period events, tremors, and volcano-tectonic earthquakes. Our method encompasses preprocessing, processing, feature extraction, feature selection, and classification stages. During the classification, we employ machine learning algorithms, specifically Decision Trees (DT), k-Nearest Neighbors (k-NN), and Support Vector Machine (SVM). The evaluation of our system performance, is assessed through the Balanced Error Rate on test data, yields significant results: 0. 1 2 for DT, 0. 1 0 for k-NN, and 0.08 for SVM. SVM algorithm presents remarkable results when applying our methodology to the feature selected matrix, which considers 29 key features, this achievement results in accuracy approaching 96% and specificity of 98%.