CONVOLUTIONAL NEURAL NETWORK APPLICATION IN VOLCANIC TREMOR IDENTIFICATION DURING 2008-2009 PRE-ERUPTIVE EPISODE OF MT. SEMERU, INDONESIA

Abstract

A convolutional neural network model has been constructed to classify volcanic tremor signals of Mt. Semeru during the year 2008 to 2009. This model was constructed by using unsupervised machine learning, precisely Convolutional Neural Network (CNN), and was trained without first providing any discriminating feature of the aforementioned volcanic tremor data. The dataset was first divided into training and testing datasets. The training dataset consists of 600 events recorded during 2008-2009 by seismic stations surrounding Mt. Semeru, including 300 volcanic tremors and 300 non-tremors. In order to increase the size of this dataset, data augmentation was carried out and producing a total of 4800 data with 2400 volcanic tremor and 2400 non-volcanic tremor events. The initial model was then trained by using 10, 15, and 20 epochs, resulting in the final model having training and validation accuracy rates of 99.51% and 96.93%, respectively. This model has successfully detected volcanic tremor signals in our testing dataset of 14 seismograms. This leads to the conclusion that unsupervised machine learning can be further utilized in volcanic activity monitoring system.