Gonca Ozata, Yavuz Selim Taspinar
In this research, two fundamental artificial intelligence applications were implemented using data from earthquakes that occurred worldwide between 2000 and 2023. The measurement type of the earthquake magnitude was classified, and the magnitude value was estimated using the regression method. The dataset used consisted of approximately 614,000 observations sourced from the United States Geological Survey. In the study, both classification and regression models were created using Artificial Neural Networks, Random Forest, Support Vector Machines, and k-Nearest Neighbour algorithms. Classification performance was evaluated using accuracy, F1 score, precision, and recall; while regression models were evaluated using metrics such as MAE, MSE, RMSE, and R². The findings show that RF and ANN achieved the highest classification performance, with overall accuracy values reaching up to ninety eight percent, whereas the SVM model demonstrated considerably lower performance. In the regression task, the RF model outperformed the other algorithms, achieving the lowest error rates and the highest explanatory power. Additionally, experimental analyses conducted across three different scenarios demonstrated that rearranging class labels in accordance with USGS standards and using feature selection based on Information Gain and Gain Ratio methods significantly increased model performance. These results demonstrate that AI-based methods can provide effective predictions in large-scale, multidimensional seismic datasets and can make valuable contributions, particularly to critical areas such as disaster risk management.
Earthquake; Classification; Regression; Prediction; Analysis