Volcanic Eruption Prediction Using LSTM and Gradient-Based Optimization Methods in Southern Peru

Quispe-Ramos J.E. · Proceedings of the 2025 IEEE 5th International Conference on Advanced Learning Technologies on Education and Research, ICALTER 2025 (2025) · DOI: 10.1109/icalter69698.2025.11355128

Resumen

In southern Peru, active volcanoes such as Ubinas and Sabancaya pose a significant threat to nearby populations due to their frequent eruptive activity. To mitigate this risk, we propose a method for forecasting volcanic eruptions in advance. This study introduces a predictive model based on Long ShortTerm Memory (LSTM) neural networks, trained using historical ground deformation data from the Global Navigation Satellite System (GNSS) and validated eruption records provided by the Geophysical Institute of Peru (IGP). Three gradient-based optimization algorithms-Adam, SGD, and RMSprop-were evaluated in conjunction with sliding time windows. The results show that the model trained with RMSprop achieved the best performance, reaching 92% accuracy and 100% recall, outperforming Adam (91%) and SGD (90%). This approach enables highly reliable eruption forecasting and represents a valuable tool for risk reduction in volcanically active regions.

Instituciones: Universidad Nacional del Altiplano