AI-POWERED EARTHQUAKE RESILIENCE: PREDICTIVE MODELING AND DAMAGE LEVEL ASSESSMENT FOR SEISMIC-RESISTANT STRUCTURES

Abstract

Assessing the seismic resilience of structures is critical for ensuring safety and minimizing economic losses during earthquakes. Traditional analysis methods, while detailed, often require significant computational resources and expertise, limiting their applicability for rapid evaluations or large-scale assessments. This study explores the potential of machine learning (ML) as an efficient alternative for predicting key seismic performance parameters. We developed predictive models using a dataset comprising building characteristics and seismic load information, including features such as Seismic Zone Classification, Peak Ground Acceleration (PGA), Building Height, Structural Type, Damper Type, Number of Storeys, Damping Ratio, and Soil Type. Random Forest algorithms were employed to build the predictive models: two Random Forest Regressors were trained to estimate Maximum Lateral Displacement (mm) and Base Shear Force (kN), and a Random Forest Classifier was trained to determine the Performance Level Classification (e.g., Immediate Occupancy, Life Safety, Collapse Prevention). Data preprocessing involved Label Encoding for categorical features and Standard Scaling for numerical features to prepare the data for model training. The models were trained and evaluated using a standard train-test split methodology. The results demonstrated the high predictive capability of the Random Forest approach for this problem. The regression models achieved excellent performance on the test set, with R-squared values exceeding 0.97 for both Maximum Lateral Displacement and Base Shear Force predictions, indicating a strong correlation between predicted and actual values. The Performance Level Classification model exhibited near-perfect accuracy (1.0 on the test set) and correspondingly high precision, recall, and F1-scores across all performance classes. To enhance accessibility and practical application, the trained models and preprocessing steps were integrated into a user-friendly web application developed using the Flask framework. This tool allows users to obtain rapid seismic performance predictions through either manual input of building/seismic parameters or batch processing via CSV file upload. This research highlights the efficacy of Random Forest algorithms for earthquake resilience prediction and provides a valuable tool for preliminary seismic assessment, potentially aiding engineers and stakeholders in design and decision-making processes.