This research paper presents an analysis of a dataset covering significant earthquakes over the past century, sourced from a publicly accessible seismic database. The dataset includes vital information such as the geographical coordinates, magnitudes, and depths of historical earthquake occurrences. The objective is to utilize machine learning techniques—specifically, k-nearest neighbors (KNN), support vector machines (SVM), random forests, and the XGBoost algorithm—to create predictive models that can anticipate future seismic events with magnitudes of 6 or higher.
The models employ latitude, longitude, and depth as input parameters to define the spatial attributes of seismic activity, while the magnitude is used as the target output parameter, reflecting the event's strength and potential destructiveness. The research encompasses rigorous data preprocessing, including cleaning and feature scaling, followed by careful model training and validation through cross-validation methods to ensure the fidelity and robustness of the predictive models. Through iterative optimization, including hyperparameter tuning, feature selection, and performance assessment via suitable evaluation metrics, the models are continuously improved. The paper focuses on detailing these processes to demonstrate the methodology behind the development of machine learning models for earthquake prediction.