Fuzzy C-Means clustering and LSTM-based magnitude prediction of earthquakes in the Aegean region of Türkiye

Türkiye is highly susceptible to earthquakes due to its active tectonic structure and the presence of major fault lines. The accurate estimation of earthquake magnitudes is essential for effective risk mitigation and structural resilience. This study proposes an integrated methodology combining clustering, statistical modeling, and deep learning techniques for the analysis and forecasting of earthquake magnitudes. Initially, earthquakes are classified into three distinct regions using the Fuzzy C-Means (FCM) clustering algorithm. For each region, statistical distributions are applied to characterize magnitude behavior. Subsequently, the Long Short-Term Memory (LSTM) model is used to predict future earthquake magnitudes. The joint application of these three methods provides a comprehensive framework for regional seismic analysis. The findings suggest that the Gumbel distribution offers the best fit for modeling return periods of earthquakes with magnitudes greater than Mwg 5.18, where Mwg denotes the global moment magnitude. Estimated return periods range from 2.56 to 3.63 years in the first region, 2.53 to 3.55 years in the second region, and 2.71-4.22 years in the third region, based on probability levels between 25 and 95%. The LSTM model forecasts that the third region is likely to experience relatively stronger seismic activity, with maximum magnitudes ranging from 2.4 to 6.5 between October 2021 and March 2029. For the same period, expected magnitudes in the first and second regions range from 2.0 to 5.7. These forecasts are supported by model performance metrics that confirm the projected magnitudes are within an acceptable and reliable range of accuracy for medium-term seismic forecasting.