Earthquake Risk Severity Prediction and Disaster Management using a Hybrid ARIMA-LSTM Model

Recent advances in earthquake prediction have rekindled interest in integrating forecasts with successful Disaster Management systems, with timescales spanning from several years to only a few months. Policymakers and disaster managers need to weigh the timing of predictions, the likelihood of false alarms, the magnitude of earthquakes, and regional susceptibility when deciding how to best prepare for seismic events, which are inherently unpredictable. Managing trade-offs between different types of errors is crucial for predictive algorithms to make effective decisions. Handling null and duplicate values was the first step in data preprocessing for this study. Then, features were extracted using IQR, CAV, and ZC as well as other methods. For the purpose of Earthquake Risk Severity Prediction, a hybrid model called ARMA-LSTM was created. This model combines AR-LSTM networks. Models were selected using the MML criterion, which achieved better results than more conventional metrics like AIC, BIC, and HQ. The AR-LSTM model that was fine-tuned had a R² score of 0.65, an MAE of 0.10, and an MSE of 0.02. These findings demonstrate the efficacy of the model and its capacity to boost catastrophe preparedness and earthquake forecasting through data- driven decision-making.