DOI: 10.29132/ijpas.1838407 ISSN: 2149-0910

A Comprehensive Benchmarking of Classical and Deep Learning Models for Global Earthquake Magnitude Prediction

Hayriye Tanyıldız, Çiğdem Ceylan, Serpil Aslan
Show PDF Cite
Earthquakes cause severe loss of life and widespread destruction, making accurate magnitude estimation a critical component of disaster management. Although literature contains numerous studies, many rely on a limited set of models or specific data types, thereby failing to provide a comprehensive and holistic comparison between classical machine learning and deep learning approaches. Furthermore, only a limited number of studies evaluate model performance using multiple and diverse error metrics, highlighting the need for a more systematic and unified analytical framework. This study presents a systematic and comprehensive comparative analysis of classical machine learning and deep learning models for earthquake magnitude estimation using a global seismic dataset covering the period from 1995 to 2023. Classical approaches such as Linear Regression, Decision Trees, Random Forest, K-Nearest Neighbor, and Support Vector Regression were evaluated alongside advanced deep learning architectures, including MLP, CNN, LSTM, BiLSTM, CNN-LSTM, CNN-GRU, and GRU. Model performances were assessed using a set of widely adopted regression-based evaluation metrics, including MAE, MSE, RMSE, R², MSLE, and Log-Cosh, supported by detailed prediction–actual comparisons, residual error distribution analyses, and time-series visualizations. Findings indicate that Random Forest significantly outperforms all other models, delivering the lowest error rates and highest explanatory power. While deep learning models offer meaningful predictions in certain contexts, they remain less effective due to the characteristics of the dataset. Overall, the study provides a rigorous benchmark and valuable reference for data-driven risk analysis, early warning systems, and machine learning-based disaster management applications.

More from our Archive