Zaman Serisi Yöntemleri ile Deprem Verilerinin Analizi

Abstract

Turkey is situated on numerous active fault lines due to its unique geographical location, and as a result, it frequently experiences earthquakes of varying magnitudes. These earthquakes have profound and often irreversible effects on human life, both economically and psychologically. To analyze the earthquake data, the study initially employed classical regression models, followed by the application of more advanced techniques such as quantile regression and the machine learning-based XGBoost regression model. From a statistical standpoint, the LSTM method was determined to be the most suitable among the methods applied. The modeling process considered earthquake magnitude (Mw), depth, and epicenter coordinates (latitude and longitude) as key variables. In the application phase of the study, analyses were particularly focused on the Marmara Region and the province of Istanbul. For Istanbul's 39 districts, earthquake magnitude predictions were generated at varying depth levels (10 and 30 km). The resulting predictions were visualized using both line graphs and heat maps, enabling a comparative assessment of regional risk distributions. Furthermore, the impact of depth on earthquake magnitude was examined statistically through linear regression analysis. The results indicated a statistically significant relationship between depth and Mw, concluding that an increase in depth tends to lead to higher predicted Mw values. Overall, the findings suggest that earthquakes exhibit certain statistical structures and spatial patterns. The developed models provide scientifically reliable forecasts for potential future earthquakes. This thesis demonstrates that it is possible to construct appropriate models for predicting earthquake magnitudes and that these models can yield robust and credible scientific estimations.