Hybrid Ensemble Models for Classifying Seismic Event Detection

Seismic event classification is vital for enhancing seismic hazard assessment and urban safety, especially in regions with frequent seismic activity. This study presents a machine learning-based framework for seismic event classification, aimed at improving seismic hazard assessment by distinguishing between earthquakes (EQs) and quarry blasts (QBs). Traditional manual methods are often inefficient and error-prone, especially for large-scale analyses, underscoring the need for automated solutions. This framework leverages ten carefully selected seismic features and compares multiple machine learning models, including CatBoost, AdaBoost, Support Vector Machine (SVM), Multilayer Perceptron (MLP), and Stacking ensembles. Notably, the Stacking model achieves the highest classification accuracy at 99.39%, outperforming individual models and underscoring the efficacy of ensemble learning in enhancing robustness and generalization. While the current model achieves high accuracy, the potential for further improvement exists by integrating additional seismic features, such as frequency-domain and spatiotemporal data. Testing with larger, varied datasets are important to check the solidness of these models across diverse geological conditions, setting a foundation for advanced automated seismic classification research.