Earthquakes are among the most destructive natural phenomena, occurring suddenly and often causing severe human and infrastructural losses. Accurate and timely prediction of significant seismic events remains a major challenge in seismology. This study presents a systematic machine-learning-based framework for earthquake prediction using historical seismic data. A three-phase methodology is adopted: (i) data acquisition and preprocessing of seismic records obtained from the United States Geological Survey (USGS), (ii) class balancing and feature preparation to address extreme class imbalance inherent in earthquake datasets, and (iii) model development and evaluation using multiple supervised learning algorithms. Binary classification was first employed to distinguish significant from non-significant earthquakes, followed by multi-class classification to categorize events into Minor, Light, Moderate, and Strong classes. Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and Neural Network models were evaluated. After balancing, the binary dataset achieved a 50–50 class distribution, while the multi-class dataset was uniformly balanced across all four classes. Experimental results demonstrate that the Random Forest model achieved the highest binary classification accuracy of 98.15%, with an Area Under the Curve (AUC) of 0.683, indicating strong discriminative capability. The Neural Network model achieved the highest recall (0.611), making it suitable for early warning scenarios where missed detections are critical. For multi-class classification, Random Forest achieved the highest overall accuracy of 87.78%, outperforming other models in robustness and stability. These results confirm the effectiveness of ensemble-based learning for seismic event prediction and highlight the trade-off between accuracy and sensitivity across different algorithms. The proposed framework provides a reliable foundation for earthquake early-warning systems in high-risk regions.
Gamal et al. (Tue,) studied this question.