Earthquake Prediction for Turkey: Ensemble Learning Approach

The prediction of earthquakes is a critical field of study in seismology and geophysics due to their substantial risks to human lives and infrastructure. The conventional approaches to earthquake prediction frequently depend on historical data and empirical models, which exhibit restricted accuracy and lead times. In recent times, the use of machine learning methodologies has surfaced as a promising approach for enhancing earthquake prediction capabilities. One of the key benefits of machine learning is its capacity to effectively evaluate extensive and heterogeneous datasets, encompassing seismic waveforms, geographical data, and historical records about earthquakes. Machine learning models can use these data sources to discern concealed patterns and connections, offering significant insights into the anticipation of earthquakes and the spatial and temporal distribution of seismic occurrences. In addition, machine learning algorithms have demonstrated potential in real-time seismic monitoring systems, facilitating the implementation of early warning mechanisms and expediting reaction techniques. These systems can offer crucial seconds to minutes of advanced warning, potentially preventing loss of life and reducing damage in locations susceptible to earthquakes. The dataset used in this study consists of earthquake data that occurred in Turkey between 1915 and 2023, taken from the Kandilli Observatory of Boğaziçi University. Extra Trees Regressor, Random Forest Regressor, Catboost Regressor, and Extreme Gradient Boosting algorithms were applied to the dataset, and the Extra Trees Regressor was given the best prediction score. The results of these algorithms were compared briefly.