From Historical to AI-Based Approaches in Earthquake Forecasting: The Case of Türkiye

Because of the presence of active fault zones, Turkey possesses a strong seismic hazard potential. The significance of machine learning and deep learning-based techniques in understanding seismic behavior is discussed in the paper. The traditional techniques followed in earthquake prediction are also discussed in the paper. The data set required for the analysis was provided by the Kandilli Observatory and Earthquake Research Institute (KRDAE). The data set contained 66,105 earthquake data from 1900 to 2024. LSTM, Random Forest and GRU are some of the seven different techniques taken into consideration while making the assessment. The performance of the models was also compared using MAE, MSE, and RMSE. The mean squared error (MSE) calculated for the Random Forest algorithm was found to be 0.1335. The results show that deep learning and artificial intelligence-based techniques are effective in simulating small and medium-sized earthquake magnitudes and also offer a meaningful approach for predicting the magnitudes, frequencies, and locations of earthquakes.