Machine learning models for seismic damage assessment in earthquake-prone regions

Abstract

The destruction of seismic structures poses a lot of risks to the life of human beings and the economic stability and the critical infrastructure of the tectonically active areas. Proper evaluation of the damage after the occurrence of an earthquake is critical in informing emergency response, ranking intervention during rescue and determining long term city rebuilding strategies. Traditional field based tests may often be sluggish, subjective and resource heavy, especially in congested or otherwise unreachable regions. The most recent developments in machine learning (ML) and remote sensing technologies provide revolutionary alternatives that can be used to create seismic damage maps automatically and quickly. The work is an overall review of ML models in seismic damage prediction with experimental analysis of hybrid networks, deep convolutional neural networks (CNNs), transformer-based image models, and multimodal fusion networks. The study relies on multivariate data, such as satellite data, LiDAR data based structural geometries, ground motion parameters and building inventories to forecast the extent of structural damage at various regions of earthquake prone areas. Analytics prove that ensemble ML techniques are more predictive in tabular engineering problems, but deep learning predictors are more efficient in identifying visual damage in tasks. The research has also found the issues associated with imbalance of data, transferability, geospatial heterogeneity, model interpretability and constraints regarding computation. This suggests a hybrid integrated architecture, that is, a combination of CNN-based feature extraction and gradient-boosting classifiers and geospatial weighting schemes, to provide regional flexibility. Comprehensively, the results demonstrate that ML is a paradigmatic change in post-earthquake risk assessment and has significant consequences to the policy of disaster management, the preparedness against earthquakes, and resilience in infrastructure design.