Earthquake prediction using deep learning and animal behavioral readings

In this study, we explore the innovative application of deep learning

models to predict earthquakes using behavioral data from farm animals, drawing inspiration from the pioneering work of Wikelski et

al. (2020). Our approach utilizes a diverse array of models, including Parallel Layer Neural Networks, Regularized Skip Neural Networks, and advanced architectures such as the Transformer models, and an

Encoder Neural Network. The Encoder Neural Network is specifically designed to process and extract meaningful features from

complex temporal animal behavior data, enhancing the predictive capabilities of our models. Custom loss functions are integrated to address the challenges unique to earthquake forecasting, such as data imbalance and the critical need to minimize false negatives. Our models are rigorously evaluated based on F1 score, precision,

and accuracy metrics. This research aims to establish a ground- breaking method for short-term earthquake prediction, potentially

revolutionizing disaster preparedness and risk mitigation. The results could significantly advance the field of seismology, harnessing

animal behavioral patterns as a natural indicator of impending seismic events.