The Application of Machine Learning Algorithms on Triaxial Passive Seismic Data to Identify The Geological Location of The Signal Source
Muhammad Randy Azhari, Bagus Mahendro Wibowo Adhi, Evi Fazriati, Yudi Rosandi (*).

Department of Geophysics, Faculty of Mathematics and Natural Sciences, Universitas Padjadjaran, Jl. Raya Bandung Sumedang km. 21 Jatinangor, Kab. Sumedang, Jawa Barat, Indonesia.
*rosandi[at]geophys.unpad.ac.id

Abstract

The characteristics of ground vibration are determined by the local geological and physical conditions of the Earth. Such vibrations can be detected using the passive seismic measurement. This research aims to create an advanced signal processing program to classify the local characteristic of ground vibration signals, applying the machine learning techniques, specifically the Convolutional Neural Network. The research process involved data acquisition, data preprocessing, model creation, model training, and model testing. Data acquisition was performed using a triaxial seismometer. The acquired data was converted into image representations in the form of axial spectrograms. The training data was divided into three directional components. The training process consists of two steps namely the component identification step and classification step. For the component identification we obtained accuracy of 87.8%. Whereas, for the classification step we obtain 90.8% using the horizontal model and 95.3% using the vertical model. Based on the confusion matrix evaluation, the model achieved an accuracy of over 85% in the overall classification. Furthermore, in the testing process correct classifications that matched the labels in all experiments was achieved. This work demonstrates the capability of classifying the local characteristic of ground vibration signals by means of the Convolutional Neural Network algorithm.

Keywords: Machine Learning, CNN, Microtremor, Geological Classification

Topic: Physic