The analysis and classification of seismic records of volcanoes allow us to determine the alert state in which it is. A timely study of these signs can contribute to decision-making to safeguard the integrity of people in the face of a natural disaster. The present work applies a methodology that combines the analysis of linear prediction coefficients and artificial neural networks to classify earthquakes. Two types of earthquakes that come from Galeras Volcano, Colombia, are studied volcano-tectonic and long-period. The classification is made using the clustering technique based on unsupervised learning. The signals are transformed using the linear prediction filter coefficients technique, which has the function of reducing the size of the vector that contains the original data. MATLAB software is used to generate a self-organizing network that handles clustering. The results show that the best alternative in unsupervised learning is to use the linear prediction coefficients of order 5, 6, and 7 to represent a seismic signal. For lower orders, the necessary information is not captured and for higher orders, noise information is shown.