Earthquakes have historically been a prominent cause of both loss of life and substantial economic damages. Türkiye, home to significant tectonic fault lines, carries a considerable earthquake risk. Therefore, adequate preparedness for earthquake risk assumes paramount importance. In this preparatory phase, having prior knowledge regarding potential earthquake magnitudes becomes crucial. This study initiates by performing statistical feature extraction on earthquake data derived from the San Andreas Fault Zone (SAFZ), a fault zone closely resembling the North Anatolian Fault Zone (NAFZ) in terms of tectonic activity. This data, sourced from the United States Geological Survey (USGS), was organized into time intervals, leading to the derivation of 15 new features.
Subsequently, an average threshold value was established, utilizing error rates associated with the feature values correlated with earthquake data from the NAFZ as weighting factors. The NAFZ dataset was divided into three groups based on this threshold value. Machine learning and deep learning network models, specifically XGBoost, Svr, and Lstm, were then applied to each dataset group to estimate the magnitude of earthquakes centered in the NAFZ. The Lstm model yielded the lowest Mean Absolute Error (MAE) and Mean Squared Error (MSE) values within the dataset group falling below the threshold value. This study effectively demonstrates the tectonic similarity between the SAFZ and NAFZ, grounded in numerical data analysis. ORCID NO: 0000-0003-3216-1979
Anahtar Kelimeler: Machine Learning, Artificial Neural Network, Time Series Data, Earthquake Magnitude Estimation, Support Vector Regression, Extreme Gradient Boosting, Long-Short Term Memory