Classification of Earthquake Vibrations Using the ANN (Artificial Neural Network) Algorithm

Abstract

The Indo-Australian Plate, the Eurasian Plate, and the Pacific Plate all converge where Indonesia is situated. As a result, Indonesia is a nation where earthquakes occur frequently. Some researchers have studied machine learning algorithms for categorizing earthquake vibrations. In this experiment, earthquake vibrations are categorized using the Artificial Neural Network method. We need appropriate datasets to obtain the maximum accuracy from the artificial neural network technique. The findings of this experiment show that feature extraction is required for the datasets to be trained to obtain a high accuracy value. The mean, median, maximum, minimum, skew, and kurtosis values are the feature that are extracted. In addition to employing feature extraction, it is crucial to modify the algorithm model. The experimental setup that uses “sigmoid” activation on the input layer, the three hidden layers, and the output layer yields the best accuracy when all feature are extracted, with training to test ratio of 90% to 10%. This is demonstrated by the exceptional training accuracy and testing accuracy values, which are 99.85 percent for training accuracy and 99.12 percent for validation accuracy. The mean value yields the highest accuracy result compared to employing just one feature extraction. Only 90.97 and 90.37 percent of training and validation accuracy are obtained when the mean is used alone for feature extraction.