Determination earthquake hazard maps with machine learning techniques

Due to its geographical conditions, our country frequently faces the reality of natural disasters, especially earthquakes. Considering that most of the loss of life and property is caused by earthquakes and this region is shaken by a devastating earthquake every five years, earthquake disaster comes first in terms of disasters that require precautions. It is of great importance to be able to predict earthquakes in order to determine the precautions that can be taken for earthquakes. In this context, earthquake prediction studies with machine learning have gained momentum in recent years. It is seen in the literature that these studies have been carried out by various methods. In this study, a data set was created in which earthquake catalogue, geological data and geodetic data were combined. Afterwards, both the raw data including all earthquakes and the data set to which fault information and strains are added, are appropriately separated as training and test data. By using Random Forest, Xgboost, Decision Tree and K-nn regression algorithms, the models were trained with the training set and the trained models were tested with the test data. Obtained results were evaluated and compared. According to the square mean error values, the best result was obtained in the random forest and xgboost algorithms with 0.09, while the worst result was obtained in the decision tree algorithm with 0.19. Finally, earthquake predictions for the future were made with all the algorithms used, and these predictions were used in the production of earthquake hazard maps. This study contributed to the literature by bringing a different perspective to the data set used in earthquake prediction studies with machine learning.