Using machine learning to predict the extent of earthquake damage to concrete buildings given the Geotechnical properties

This paper aims to predict the damages of earthquakes to concrete buildings by considering the hardness of the soil where the buildings are constructed. Three methods were used: k-nearest neighbor, random forest, and support vector machine (SVM) classifier decision tree method in Python. The dataset consists of ۱۰۰ concrete buildings from around the world that were damaged by different earthquakes. The dataset include the magnitude, depth, duration, and distance of the earthquake, as well as the soil hardness according to the USDA soil taxonomy. The level of damage to buildings was grouped into six ranks from no damage to collapse. The results showed that the random forest method predicted the damages with ۶۸ percent accuracy, which was better than the other two methods. This paper demonstrates the potential of using machine-learning techniques to assess the seismic vulnerability of concrete buildings.

Keywords:

earthquake damage prediction , concrete buildings , soil hardness , machine learning , SVM