Seismic performance prediction of reinforced concrete buildings using the RYTEIE method: a comparison of machine learning models

Abstract

This study aims to predict seismic performance scores defined under the “Principles for the Identification of Risky Structures” using a final dataset of 3,543 buildings, obtained after preprocessing raw data from 4,200 reinforced concrete buildings in central Elazığ. Various supervised machine learning models were compared. Input parameters included building identification data, geometric characteristics, and structural irregularities, while the seismic performance score was the output. Models such as KNN and Random Forest were trained and evaluated using regression metrics including R-squared (R²), mean absolute error (MAE), and mean squared error (MSE). Results indicate that the Random Forest model predicts seismic performance scores with high accuracy. These findings suggest that such approaches can serve as fast, effective, and reliable decision-support tools for post-earthquake building risk prioritization. Additionally, they help reduce time and labor costs in field data collection, contributing to more efficient disaster management and improved urban resilience. The study is expected to support urban transformation and disaster management strategies in earthquake-prone regions like Elazığ.