Automatic Seismic Event Classification Using Transformer Models

In this paper, we propose an automatic seismic event classification method based on transformer models to maximize the classification accuracy of volcanic raw signals. The main contribution is to develop a model that efficiently captures temporal dependencies in seismic signals through advanced machine learning architectures. The proposed method was trained and validated on the publicly available MicSigv1 dataset from the Cotopaxi volcano. The best model, using a ten-fold cross-validation strategy, achieved mean F1-score, accuracy, precision, recall, and area under the receiver operating characteristic curve (AUC) scores of 0.959, 0.959.5, and 0.966, respectively, in the training phase and in the test phase. The successful classification performance across multiple metrics highlights the effectiveness of the proposed method in capturing temporal and spatial patterns in raw seismic signals to maximize classification without using the traditional workflow involving feature calculation or feature space transformation, as seen in some past approaches.