Feature-Driven Earthquake Forecasting: From Seismic Data to Early Warning Potential
Faculty Mentor
Lei Chen
Location
Russell Union Room 2080
Type of Research
Published
Session Format
Oral Presentation
College
Allen E. Paulson College of Engineering & Computing
Department
Applied Computing
Abstract
Accurate short-term earthquake forecasting remains one of the most challenging problems in seismic risk management. This study presents a machine learning–based framework for forecasting the maximum expected earthquake magnitude range in Los Angeles within a 30-day window. Using historical seismic records, a comprehensive feature matrix was constructed by integrating established seismological indicators with newly engineered predictive variables to enhance temporal and magnitude sensitivity.
Sixteen machine learning algorithms were systematically evaluated to determine optimal classification performance. Structured preprocessing, feature selection, and model validation techniques were applied to improve generalization and reduce overfitting. The proposed framework predicts the maximum expected earthquake magnitude category within predefined magnitude ranges. Among the evaluated models, the Random Forest algorithm demonstrated superior performance, achieving approximately 98% classification accuracy in identifying the correct magnitude range.
The results indicate that carefully engineered feature subsets combined with ensemble learning methods can significantly enhance short-term magnitude range forecasting. This approach advances data-driven seismic hazard assessment and supports more informed preparedness, mitigation planning, and risk management strategies in earthquake-prone urban regions. By integrating artificial intelligence with geophysical analysis, this research contributes to the growing interdisciplinary field of computational earthquake forecasting.
Program Description
.
Start Date
4-23-2026 10:00 AM
End Date
4-23-2026 10:15 AM
Recommended Citation
Yavas, Cemil Emre, "Feature-Driven Earthquake Forecasting: From Seismic Data to Early Warning Potential" (2026). GS4 Student Scholars Symposium. 9.
https://digitalcommons.georgiasouthern.edu/research_symposium/2026/2026/9
Feature-Driven Earthquake Forecasting: From Seismic Data to Early Warning Potential
Russell Union Room 2080
Accurate short-term earthquake forecasting remains one of the most challenging problems in seismic risk management. This study presents a machine learning–based framework for forecasting the maximum expected earthquake magnitude range in Los Angeles within a 30-day window. Using historical seismic records, a comprehensive feature matrix was constructed by integrating established seismological indicators with newly engineered predictive variables to enhance temporal and magnitude sensitivity.
Sixteen machine learning algorithms were systematically evaluated to determine optimal classification performance. Structured preprocessing, feature selection, and model validation techniques were applied to improve generalization and reduce overfitting. The proposed framework predicts the maximum expected earthquake magnitude category within predefined magnitude ranges. Among the evaluated models, the Random Forest algorithm demonstrated superior performance, achieving approximately 98% classification accuracy in identifying the correct magnitude range.
The results indicate that carefully engineered feature subsets combined with ensemble learning methods can significantly enhance short-term magnitude range forecasting. This approach advances data-driven seismic hazard assessment and supports more informed preparedness, mitigation planning, and risk management strategies in earthquake-prone urban regions. By integrating artificial intelligence with geophysical analysis, this research contributes to the growing interdisciplinary field of computational earthquake forecasting.
