Earthquake Occurrence Prediction in Türkiye Using Machine Learning Models: A Rare-Event Classification Approach

Hardal, Yunus Emre (2025) Earthquake Occurrence Prediction in Türkiye Using Machine Learning Models: A Rare-Event Classification Approach. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (900kB) | Preview
Preview	PDF (Configuration Manual) Download (766kB) | Preview

Abstract

Turkey is among the most active seismic areas in the world and has a continuous occurrence of low-intensity earthquakes along with occasional destructive events. The need for reliable, data-driven predictive tools is apparent. This study investigates two machine-learning methodologies that can be used together to create effective predictive tools, which utilize a large dataset of 493,000 seismic events. One was to classify earthquakes into two categories of destructive and non-destructive and one was to classify earthquakes into 5 magnitude classes (ML, Md, Ml, Mw, MW) using principal components of seismic attributes. Due to the significant imbalance of the destructive event class, the models did not require any’s balancing techniques within the training sets because the scientific community understands that the destructive earthquake events are difficult to predict. As a result, despite the fact that all models displayed exceptional overall accuracy, they missed all minority destructive cases, resulting in zero F1 score.

The second task, magnitude type classification, was completed without including the rare event classes to create valid stratified splits for evaluation purposes and for classifying ML, Md, Ml, Mw and MW, respectively. The training data consisted of four numeric parameters (lat/lon/depth/magnitude). Logistic Regression, KNN, Random Forest, Gradient Boosting, AdaBoost and XGBoost were among several models to benchmark. The majority of the classes ML and Md produced excellent predictive capabilities, whereas the remaining classes were difficult to predict. XGBoost provided the highest accuracy score (0.8591). Also, the model improved balance across all classes to some extent.

These findings demonstrate both the power and the limitations of machine learning in applying seismic prediction modelling: while ML can be successfully applied to predicting the magnitude type of an earthquake, applying this methodology to predict a disaster event is much more difficult due to the limited availability of cases and the natural unpredictability of seismic occurrence.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Zahoor, Sheresh UNSPECIFIED
Subjects:	G Geography. Anthropology. Recreation > GE Environmental Sciences G Geography. Anthropology. Recreation > GE Environmental Sciences > Earth sciences > Geology > Physical geology > Sedimentation and deposition > Earth movements > Earthquakes Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	07 Sep 2026 11:46
Last Modified:	07 Sep 2026 11:46
URI:	https://norma.ncirl.ie/id/eprint/9863

Actions (login required)

View Item