Increasing tne effectiveness of rapid risk detection methods in reinforced concrete buildings by using machine learning methods

Turkey is located in a high seismic hazard zone, and a significant portion of its existing building stock is inadequate to withstand earthquakes. The 6 February 2023 Kahramanmaraş earthquakes demonstrated that the rapid and reliable identification of risky buildings is critical for disaster management. The aim of this thesis is to improve the effectiveness of rapid risk detection methods used in reinforced concrete buildings through machine learning methods and to improve post-earthquake building seismic risk prioritisation processes. The hypothesis is that machine learning models developed using field data and official institutional data separately will provide high accuracy. Two independent datasets were used in the study. The first dataset consists of field data from 4,200 reinforced concrete buildings obtained through fieldwork conducted in the central neighbourhoods of Elazığ Province. Using these data, performance scores were calculated using FEMA-154, Principles for the Identification of Risky Structures (RYTEİE), Canadian Seismic Screening, and Indian rapid visual assessment methods, and modelled using machine learning algorithms (XGBoost, Random Forest, KNN, AdaBoost). The second dataset consists of official data on post-earthquake building damage obtained from the Ministry of Environment, Urbanisation and Climate Change. This dataset was also processed using the same method, and separate machine learning models were trained and their accuracy performance was compared. The results showed that the XGBoost and Random Forest algorithms provided the highest accuracy rates in both datasets and significantly improved prediction performance compared to traditional rapid assessment methods. The model outputs showed high consistency with field observations and post-earthquake actual damage data. In particular, variables such as number of floors, building age, presence of soft floors, and material strength were found to play a decisive role in the models. This approach offers important advantages in terms of rapid applicability, low cost, large-scale assessment capabilities, and effective management of post-disaster resources for both field data and official data, thereby directly contributing to the enhancement of urban resilience.