aspg-28-3861-1792156487 20261016131447 American Scientific Publishing Group admin@americaspg.com American Scientific Publishing Group Journal of Artificial Intelligence and Metaheuristics JAIM 2833-5597 2025 9 2 Nima Khodadadi Department of Civil and Architectural Engineering, University of Miami, Coral Gables, FL, USA Earthquakes are among the most destructive natural hazards, causing severe damage to communities, loss of life, and major economic consequences. This study evaluates machine-learning models for magnitude prediction using an open-access global earthquake dataset from 2023. Five regressors are compared: Light Gradient Boosting Machine (LightGBM), Support Vector Regression (SVR), k-Nearest Neighbors (KNN), Ridge Regression, and Extra Trees. Training uses stratified cross-validation and hyperparameter optimization. The evaluation covers Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Bias Error (MBE), the coefficient of determination (R2), Nash–Sutcliffe Efficiency (NSE), Willmott Index (WI), Pearson’s correlation coefficient (r), and Relative RMSE (RRMSE). LightGBM performs best in the reported comparison, attaining MSE = 0.0474 and R2 = 0.9241. Its leaf-wise tree construction, scalability, and regularization support generalization to unseen data while maintaining computational efficiency. The results indicate that LightGBM can capture patterns in high-dimensional seismic data and provide a reproducible analytical framework for earthquake-prone regions. In combination with other advanced ML systems, such models may contribute to early-warning research, faster emergency response, and better-informed risk planning. The use of an open dataset and documented modeling steps also supports transparency and collaborative research. 2025 41 47 3861 https://creativecommons.org/licenses/by/4.0/ 10.54216/JAIM.090205 https://www.americaspg.com/journal/28/article/3861 https://www.americaspg.com/storage/articles/manuscripts/01M4Y6DD3KWATD0F6QGXRFEH6X.pdf