LSTM-Based Models for Earthquake Forecasting on Three Boundaries of the Caribbean Plate

Earthquake forecasting methods are investigated mainly for its impact in reducing seismic risk. The objective of the present work is to investigate deep learning models based on new and published LSTM architectures. The models are implemented on multivariate time series earthquake data. The datasets are selected from three different plate boundaries of the Caribbean to assess the robustness of the models for this region. The four datasets used straddle the countries of the Northern (Transform) Boundary, Eastern (Subduction) Boundary and the Dextral (Transform) Boundary. This work establishes a baseline for future investigations of time series forecasting on LSTM-based models for earthquake data. Keywords: