Application of LSTM-based deep learning models for seismic forecasting of a potential major earthquake using geophysical techniques in the marmara Sea Region, Türkiye

This study proposes a comprehensive framework integrating state-of-the-art geophysical analytics and powerful machine learning paradigms to address seismic activity prospecting in the Marmara region, one of the most seismically active areas related to the dextral North Anatolian Fault (NAF). By leveraging Long Short-Term Memory (LSTM) networks and Gated Recurrent Units (GRU), the aim of the study is to enhance the precision of seismic event predictions and seismic hazard evaluation. Clustering algorithms and Principal Component Analysis (PCA) are also applied to historical seismic data from multiple sources to delineate seismic clusters and regions of stress accumulation in the study. The results show considerable stress build-up in central Marmara especially around Istanbul and the Princes' Islands, highlighting these areas as high-risk zones for major (M ≥ 7) earthquakes for the next 30 years. Conversely, stress release zones were only recorded mostly in western Marmara Sea and near the Thrace region (Trakya) where the accumulated stress as a result of past seismic events was released. An important seismic gap was revealed in the central Marmara region, and surrounding the Princes' Islands and the transition region of Eastern Tekirdağ and Western Istanbul, implying that these segments of the Marmara- Sea earthquake could be future sources of significant contact for separating the Baykara and Marmara main fault blocks due to accumulated stress. To bridge this seismic gap, the study introduces a holistic geophysical approach alongside machine learning models that categorize fault stress levels across the locality by utilizing significant seismic characteristics (namely Peak Ground Acceleration (PGA), Peak Ground Velocity (PGV), Peak Ground Displacement (PGD), seismic events duration, energy (Arias Intensity) and dominant frequency). The study highlights the necessity of careful monitoring, strong construction codes, and education for the general public to reduce earthquake hazards. The combination of historical data analysis, geophysical modeling and machine learning approaches is an essential step toward fully understanding the tectonic evolution of the Marmara region and contributes to seismic hazard characterization and mitigation recommendations.