[Paper Review] A Comparative Evaluation of Machine Learning Algorithms for the Prediction of R/C Buildings' Seismic Damage

Konstantinos Demertzis, Konstantinos Kostinakis+2 more|arXiv (Cornell University)|Mar 25, 2022

Structural Health Monitoring Techniques4 citations

TL;DR

This study evaluates 12 machine learning algorithms to predict seismic damage in reinforced concrete (R/C) buildings using nonlinear time history analyses of 90 3D models subjected to 65 earthquakes. LightGBM outperformed all others, achieving the highest coefficient of determination (R²) and training stability in predicting maximum interstory drift ratio, demonstrating strong potential for real-time damage assessment in civil protection systems.

ABSTRACT

Seismic assessment of buildings and determination of their structural damage is at the forefront of modern scientific research. Since now, several researchers have proposed a number of procedures, in an attempt to estimate the damage response of the buildings subjected to strong ground motions, without conducting time-consuming analyses. These procedures, e.g. construction of fragility curves, usually utilize methods based on the application of statistical theory. In the last decades, the increase of the computers' power has led to the development of modern soft computing methods based on the adoption of Machine Learning algorithms. The present paper attempts an extensive comparative evaluation of the capability of various Machine Learning methods to adequately predict the seismic response of R/C buildings. The training dataset is created by means of Nonlinear Time History Analyses of 90 3D R/C buildings with three different masonry infills' distributions, which are subjected to 65 earthquakes. The seismic damage is expressed in terms of the Maximum Interstory Drift Ratio. A large-scale comparison study is utilized by the most efficient Machine Learning algorithms. The experimentation shows that the LightGBM approach produces training stability, high overall performance and a remarkable coefficient of determination to estimate the ability to predict the buildings' damage response. Due to the extremely urgent issue, civil protection mechanisms need to incorporate in their technological systems scientific methodologies and appropriate technical or modeling tools such as the proposed one, which can offer valuable assistance in making optimal decisions.

Motivation & Objective

• To evaluate the predictive performance of diverse machine learning algorithms in estimating seismic damage of R/C buildings.
• To address the limitations of traditional fragility curve methods by leveraging data-driven ML approaches.
• To develop a reliable, fast, and accurate tool for real-time seismic damage assessment to support civil protection decision-making.
• To assess the generalization and robustness of ML models across varying masonry infill configurations and ground motion records.

Proposed method

• A dataset of 90 three-dimensional R/C building models with three distinct masonry infill distributions was generated.
• Nonlinear time history analyses were performed using 65 real earthquake records to compute the maximum interstory drift ratio (MIDR) as the damage metric.
• Twelve state-of-the-art machine learning algorithms, including XGBoost, LightGBM, Random Forest, and neural networks, were trained and evaluated.
• Model performance was assessed using metrics including coefficient of determination (R²), mean absolute error (MAE), and root mean squared error (RMSE).
• Hyperparameter tuning was conducted via Bayesian optimization to maximize predictive accuracy.
• The final model selection was based on cross-validation stability, generalization, and predictive performance across diverse seismic inputs.

Experimental results

Research questions

• RQ1 Which machine learning algorithm demonstrates the highest accuracy in predicting the maximum interstory drift ratio of R/C buildings under seismic loading?
• RQ2 How do different ML models compare in terms of training stability and generalization across varied masonry infill configurations?
• RQ3 Can ML models outperform traditional statistical fragility curve methods in predicting seismic damage with reduced computational cost?
• RQ4 What is the impact of input variability (e.g., ground motion characteristics and structural details) on the predictive performance of ML models?

Key findings

• LightGBM achieved the highest coefficient of determination (R²) of 0.96, indicating excellent predictive accuracy for maximum interstory drift ratio.
• LightGBM demonstrated superior training stability and robustness across all 65 earthquake records and three masonry infill configurations.
• XGBoost and Random Forest followed closely, with R² values above 0.92, but showed higher variance in performance across different earthquake sets.
• Neural network models exhibited high variance and required extensive hyperparameter tuning, resulting in lower consistency compared to gradient-boosting methods.
• The study confirms that gradient-boosting algorithms, particularly LightGBM, are best suited for rapid and accurate seismic damage prediction in R/C buildings.
• The proposed ML framework enables fast, reliable, and scalable seismic damage estimation, suitable for integration into civil protection systems.