Machine Learning Approach to Analyze the Impact of Seismic Force

Abstract

The study investigates the impact of seismic forces on structural response using a machine learning (ML)–based approach. The study has utilized scaled structural models and subjected them to control shaking table excitations representing low, medium, and high earthquake intensities. A linear regression–based ML model was developed using multiple input parameters to predict the structural response characteristics. The proposed model successfully captured the overall behavioral trends of the structures, achieving high validation accuracy. However, prediction deviations increased at higher peak ground acceleration (PGA) levels due to nonlinear seismic effects. Residual and root mean square error (RMSE) analyses indicate that, although the ML model slightly underestimated peak responses, it remained effective in identifying general response patterns. Comparative validation demonstrated a reasonable variation between lab observations and ML predictions, supporting the use of ML as a complementary predictive tool. The findings highlight the effectiveness of ML frameworks in achieving high prediction accuracy and their potential contribution to seismic design optimization.