Earthquake probability prediction with decision tree algorithm: The example of İzmir, Türkiye

Abstract

This study investigates earthquake records in the Izmir province of western Türkiye, focusing on seismic activity prediction through the application of decision tree models. Utilizing earthquake data from 1900 to 2024, including magnitude, depth, latitude, and longitude variables, the aim is to estimate future seismic events in a region known for its significant earthquake risks. The decision tree model, a machine learning approach, was trained with 80% of the dataset and tested on the remaining 20%. Performance was assessed using metrics such as precision, recall, F1 score, and overall accuracy, with the model achieving an accuracy rate of 92%. However, its ability to predict larger earthquakes was hindered due to the limited availability of data for highermagnitude events. A chi-square test demonstrated a statistically significant relationship between earthquake depth and magnitude. Additionally, a risk analysis map was created using Geographic Information Systems (GIS), highlighting fault lines and areas prone to frequent seismic activity. The study concludes that while the decision tree model is effective for predicting smaller earthquakes, the accuracy for larger events could be improved with more comprehensive data. These findings underscore the importance of targeted earthquake preparedness in Izmir, particularly in coastal areas susceptible to both seismic events and secondary hazards like tsunamis