Deep Learning for Predictive Analysis of Earthquake-Resistant Buildings

Abstract

Both the frequency and intensity of earthquakes are on the rise, which highlights the crucial need for predictive models to improve the resilience of buildings. Through the utilisation of a methodical approach that incorporates dimensionality reduction, recursive feature elimination (RFE), and feedforward neural networks (FFNNs), this research investigates the potential for deep learning to be utilised in the prediction of the seismic resilience of buildings. In order to preprocess the high-dimensional structural and material datasets, dimensionality reduction is utilised. This helps to streamline the feature space while preserving essential information. After that, RFE is utilised for the purpose of feature selection, with the objective of giving priority to the most relevant variables that have an impact on earthquake resistance. These variables include material qualities, structural design parameters, and geographic classifications of seismic zones. FFNNs, which exhibit robust prediction skills and adaptability to complicated, non-linear relationships that are inherent in the data, are utilised in the process of performing the final classification. According to the findings, the categorisation accuracy is quite high.