Predicting Earthquakes with Ionospheric Data: A Hybrid Approach Utilizing Deep AutoEncoder and LSTM Networks

An integral part of the Earth's atmosphere is driven by the ionosphere. Solar flares induce ionosphere anomalies as a result of coronal mass ejection, seismic activity, and geomagnetic activity. Total Electron Content is the primary metric used to study the ionosphere's structure (TEC). GPS-derived TEC values are useful for examining how the ionospheric response to earthquakes is affected. In order to identify earthquakes, this article examines the relationships between TEC data and earthquakes. Our aim is to suggest a classification strategy for identifying earthquakes that occurred in earlier days. This research discusses the ionospheric variability during moderate and severe earthquake events of varied intensity for the years 2012-2019. Deep Autoencoders are used by the suggested model to extract features from TEC data. A Stacked LSTM model was constructed using the features gathered to forecast the earthquakes that occurred in the preceding days. For evaluation, the suggested hybrid model is compared with the Support Vector Machine (SVM) and Linear Discriminant Analysis (LDA) classifier models. According to the findings, the suggested hybrid model increases earthquake detection with an accuracy rate of roughly 0.84 and is a useful tool for identifying earthquakes based on prior days.