Machine Learning for Fully Automated Detection of Volcanic Seismic Signals in Real Seismic Records at Sinabung Volcano, North Sumatra

Abstract

To mitigate damage from volcanic disasters, experts conduct thorough monitoring mainly through observing volcano earthquake types, as these patterns provide crucial insight into magma movement. Traditionally, the classification of earthquake types has been performed manually, a process that is both time-consuming and subjective. To solve this problem, several studies have developed machine-learning models to classify volcano earthquake types automatically. However, previous research typically trains and evaluates machine learning models on existing datasets. Although these models can classify volcanic earthquake types, they require manual selection of input earthquake signals for classification. To fill this gap, this study aims to develop a machine learning model that integrates with an event detection algorithm to enable fully automated detection in actual seismic recordings at Sinabung Volcano. This study employs the Short-Term Average/Long-Term Average (STA/LTA) method, which calculates the ratio between two time windows for event detection. Two machine learning models, namely a Multi-Layer Perceptron (MLP) based on neural networks and a Random Forest (RF) based on decision trees, are used to classify the events detected by the STA/LTA method. Consequently, this approach enables the machine learning models to operate fully automated. In this study, we first detect events using the STA/LTA method on a daily basis; subsequently, each detected event is classified using a machine learning model developed based on the dataset. RF and MLP successfully predict relative low difference percentage compared to the actual number in earthquake catalogue, with values of VT 5.31%, LF 46.62%, and EXs 30.95%. Automated detection and classification can improve the efficiency of mitigating the risks associated with volcanoes by identifying potential anomalies in advance.