Enhancing Earthquake Prediction using Machine Learning with Optimization Method

Abstract

Earthquakes have complex seismic characteristics that make accurate classification challenging. This study proposes a machine learning approach that combines K-Means clustering and Random Forest to improve earthquake classification performance. We utilize earthquake records with numerical and geographic attributes, including date, latitude, longitude, magnitude, depth, and location. K-Means clustering groups the earthquake records into three clusters, interpreted as Dangerous, Medium, and Not Dangerous based on their seismic characteristics. We then apply Random Forest to classify the earthquake records and evaluate different numbers of trees using the F1-Score. The results show that Random Forest with 100 trees achieves the highest F1-Score of 87.21%, slightly outperforming SVM (87.11%), Decision Tree (86.24%), and Naïve Bayes (85.89%). We further analyze several feature combinations to identify the most informative attributes. The combination of Date, Latitude, Longitude, Magnitude, and Location produces the highest F1-Score of 93.23%, improving the performance by 6.02 percentage points compared with the initial Random Forest configuration. The results demonstrate that appropriate feature selection substantially improves earthquake classification performance. This study shows that integrating K-Means clustering, feature selection, and Random Forest provides an effective approach for earthquake classification based on seismic and geographic characteristics.