Multiple Regression-Based Prediction Method to Assess the Impact of PGA and Distance on Post-Earthquake Structural Damage Levels

Authors : Ömer Faruk Nemutlu

Pages : 37-51

Doi:10.46810/tdfd.1605168

View : 76 | Download : 43

Publication Date : 2025-06-27

Article Type : Research Paper

Abstract :This study was carried out to evaluate the accuracy of the damage assessments made after the 06 February 2023 Kahramanmaraş earthquakes and to ensure that these data are a guide for future studies in the field of earthquake engineering. The relationship between damage levels, maximum ground acceleration (PGA) values measured by Disaster and Emergency Management Affair (DEMA) stations and distances to earthquake-affected cities were analyzed. Unlike the studies in literature, evaluation was made on multiple input and multiple output parameters, and a separate regression model was used for each output data. As a result of regression analysis, a significant relationship was found between damage levels and PGA-distance parameters. The R² scores for the \\\"No damage\\\" and \\\"Heavy damage\\\" levels were found to be 0.75 and 0.71, respectively. In the analyzes made by reducing the damage levels to two main categories (damaged and undamaged), the R² scores were calculated as 0.63 and 0.6, respectively. These results show that there is a sufficient level of agreement between the input and output parameters, but they reveal that the dataset should be expanded, and the positional details of the structures should be obtained separately for higher accuracy. Within the scope of the study, linear regression, polynomial regression, random forest and gradient boosting models were used and their performances were compared. According to the results obtained, gradient boosting and random forest models were the models that exhibited the best compatibility according to damage levels. In particular, the fact that the random forest model gives the best results in 5 out of 6 damage levels shows that the model is a method that produces fast and reliable results in such complex analyses. As a result, it was determined that model performance at low conforming damage levels could be improved by expanding the data set and increasing the available data details. These findings make important contributions to the accuracy analysis of damage assessments after earthquakes and provide a scientific basis for similar studies.
Keywords : Çoklu Regresyon, Yapay Zeka, Hasar Dağılımı, Deprem Gözlemi, Kahramanmaraş Depremleri

ORIGINAL ARTICLE URL