Loading...

Regional-Scale Prediction of Seismic Retrofit Method and Cost of Buildings

Najafabadi, Mohammad | 2026

77 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58868 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Bakhshi, Ali
  7. Abstract:
  8. The present study aimed to develop two machine learning–based models to predict the best seismic retrofitting method for buildings and to estimate the associated implementation costs. Seismic retrofitting is one of the most effective approaches for enhancing structural resilience and reducing vulnerability in high-seismicity areas, such as metropolitan Tehran. In vulnerable cities, retrofitting is carried out with the involvement of municipalities and relevant authorities; therefore, planning and financing large-scale retrofitting projects at the city or regional level require an accurate model for cost prediction. Previous studies in this field have been conducted with diverse approaches and datasets. This diversity includes considerations of the geographical location of structures, the use of different data sources, modeling of various structural types (e.g., concrete, steel, or masonry structures), and the application of different predictive models such as linear models or artificial neural networks. However, a significant portion of these studies did not address the retrofitting method itself or the effects of common structural deficiencies, such as poorly performing connections. This study aimed to fill this gap by developing machine learning–based models that leverage structural features to predict both the best retrofitting method and the cost of its implementation. For this purpose, after identifying structural deficiencies and factors affecting seismic performance, the input variables for each model were determined separately. The first model was then developed to predict the best retrofitting method, while the second model was developed using its own specific variables to predict retrofitting costs. To evaluate the performance of the developed models, a set of quantitative metrics was employed. In predicting the best retrofitting method, two classification models based on CatBoost and XGBoost algorithms were developed. Evaluation results indicated that the CatBoost model, with an accuracy of 0.84 and an F1-Macro score of 0.83, outperformed the XGBoost model, which had an accuracy of 0.80 and an F1-Macro score of 0.78. Consequently, CatBoost was selected as the top-performing model for predicting the retrofitting method. For predicting retrofitting costs, three regression models based on CatBoost, XGBoost, and Random Forest were developed. The CatBoost model achieved an R² of 0.73, a mean absolute error of 0.20, a root mean squared error of 0.24, and an R² of 0.87 on the training data. The XGBoost model had an R² of 0.72, a mean absolute error of 0.19, a root mean squared error of 0.25, and an R² of 0.70 on the training data. The Random Forest model achieved an R² of 0.76, a mean absolute error of 0.19, a root mean squared error of 0.23, and an R² of 0.81 on the training data. Given the high accuracy of the Random Forest model and the closeness of R² values between training and evaluation data , which indicates the absence of overfitting, this model was selected as the best option for predicting retrofitting costs. The results of this study can assist engineers, urban planners, and policymakers in decision-making regarding the prioritization of building retrofitting and in estimating the financial resources required to implement incentive-based retrofitting programs
  9. Keywords:
  10. Seismic Retrofit ; Machine Learning ; Predictive Model ; Decision Making ; Resilience Index ; Earthquake Resistance ; Predictive Model for Retrofitting Cost ; Regional Scale Seismic Retrofitting

 Digital Object List

 Bookmark

No TOC