Continuous Learning for Anomaly Detection: A Case Study in Volcanic Unrest Monitoring

Abstract

In the field of volcanology, timely detection of anomalies is essential for disaster mitigation. Traditional methods often fall short in adapting to evolving volcanic behavior. We propose a model that combines continual learning and autoencoders to adaptively detect anomalies. The autoencoder extracts relevant features from sensor data, while continual learning enables the model to adapt to changing volcanic patterns. A case study demonstrates its effectiveness in real-time monitoring, offering a data-driven and efficient solution for volcanic anomaly detection.