Earthquake Magnitude Detection: Leveraging Machine Learning with Metaheuristics for Feature Selection and Dimensionality Reduction

Abstract

The rapid and accurate estimation of earthquake magnitude is a cornerstone of effective early warning and hazard mitigation strategies. This study systematically evaluates the efficacy—Particle Swarm Optimization (PSO), Grey Wolf Optimizer (GWO), and Whale Optimization Algorithm (WOA)—as feature selection wrappers for improving earthquake magnitude prediction. Four distinct ML models were benchmarked: Random Forest (RF), Light Gradient Boosting Machine (LightGBM), Support Vector Machine (SVM), and Decision Tree (DT). These models were trained and evaluated on a seismic dataset comprising 1000 events characterized by 598 features. Model performance was rigorously assessed using Mean Squared Error (MSE), Mean Absolute Error (MAE), the Coefficient of Determination ($R^2$), feature reduction rates, and total execution time. The results demonstrate that integrating a metaheuristic feature selection approach can yield substantial performance gains. The PSO-optimized Random Forest model achieved the best overall performance, with an MSE of 0.0284 and an $R^2$ of 0.8547, representing a 23.8 % improvement in MSE over its baseline. Feature selection proved to be a critical enabler for the SVM, transforming it from a non-performing model ($R^2$= −0.1716) into a viable predictor, with the PSO-optimized variant achieving an $R^2$ of 0.6517. The findings conclude that a PSO-based feature selection wrapper offers a robust and highly effective method for enhancing the accuracy and computational efficiency of advanced ML models for seismic analysis.