Seismic Performance Assessment of Reinforced Concrete Building Stock Using Artificial Neural Network and Linear Regression Analysis

Istanbul is located on extensive piece of land which is susceptible by seismic activity. In the last half century, Turkish earthquake codes for designing building under earthquake loads went through many modifications and editions (TEC1975, TEC1997, TEC2007, and TBEC2018). Hence, there are many buildings existing that has been built in accordance with old regulations since improvements in the recent earthquake code. Therefore, the need of a quick assessment method to identify the building seismic performance level in accordance with the latest seismic code is extremely vital. For this purpose, this research is aiming to prepare a database for the quick estimation on building seismic performance by constructing an artificial neural network model that is capable of this, relating building material properties, geometry, designed standard, site class, and peak ground acceleration to the building seismic performance levels. In order to meet these objectives, 540 reinforced concrete building models with various parameters are modeled with respect to TEC1975, TEC1997, TEC2007, TBEC2018 and seismic performance obtained from the analysis in accordance with TBEC2018. Data obtained are used to train and validate the constructed artificial neural network (ANN) model. Also, several training algorithms performed with various number of hidden layers and comparison between them is discussed in order to figure out the optimum number of hidden layers and best train method which gives the highest accuracy of prediction for the performance assessment of the buildings. Since the artificial neural network model created for the performance level estimation of the existing buildings, validity of the created model is checked by the application through the existing buildings as a case study with various parameters within the range of considerations according to the existing study. The data obtained from the analysis is used to perform multiple linear regression analysis (MVLRA) as well. Results indicate that ANN can be a very profound technique in predicting the seismic performance levels with a determination coefficient (R2 ) of 0.8786. Furthermore, identification of the significance of the predictor variables according to their effect on seismic assessment have been done with several methods which are widely used in literature as well.