Bayesian Optimized LightGBM model for predicting the fundamental vibrational period of masonry infilled RC frames

Abstract

The precise prediction of the fundamental vibrational period for reinforced concrete (RC) buildings with infilled walls is essential for structural design, especially earthquake-resistant design. Machine learning models from previous studies, while boasting commendable accuracy in predicting the fundamental period, exhibit vulnerabilities due to lengthy training times and inherent dependence on pre-trained models, especially when engaging with continually evolving data sets. This predicament emphasizes the necessity for a model that adeptly balances predictive accuracy with robust adaptability and swift data training. The latter should include consistent re-training ability as demanded by real-time, continuously updated data sets. This research implements an optimized Light Gradient Boosting Machine (LightGBM) model, highlighting its augmented predictive capabilities, realized through the astute use of Bayesian Optimization for hyperparameter tuning on the FP4026 research data set, and illuminating its adaptability and efficiency in predictive modeling. The results show that the R² score of LightGBM model is 0.9995 and RMSE is 0.0178, while training speed is 23.2 times faster than that offered by XGBoost and 45.5 times faster than for Gradient Boosting. Furthermore, this study introduces a practical application through a streamlit-powered, web-based dashboard, enabling engineers to effortlessly utilize and augment the model, contributing data and ensuring precise fundamental period predictions, effectively bridging scholarly research and practical applications.