Evaluation of Concrete Performance of Risky Buildings in Earthquakes using Machine Learning

Abstract

Predicting the strength of building materials is nowadays a vital process for taking precautions against earthquakes. Increasing the efficiency of this process by reducing its cost can prevent possible loss of life. In this study, machine learning based models were used. First, vibration data were collected on 6 cylinder and 33 standard cube concrete specimens using accelerometers and piezoelectric sensors, followed by preprocessing steps such as noise reduction and normalization. XGBoost and Random Forest regression algorithms were used in the model training process. The cross-validation value MAE (Mean Absolute Error) obtained from the XGBoost model trained with data from cube samples was 6.45, compared to the MAE of 10.9 obtained from the Random Forest model. However, due to the limited data set, it is predicted that there may be variability in the results of the models with larger data sets.