What question did this study set out to answer?

The aim is to develop a machine learning framework that efficiently assesses seismic damage and estimates fragility curves following earthquakes.

July 8, 2026

Physics-Aware Interpretable Machine Learning for Rapid Seismic Damage Screening and Fragility Curve Estimation Using Post-Earthquake Building Survey Data

Key Points

The aim is to develop a machine learning framework that efficiently assesses seismic damage and estimates fragility curves following earthquakes.
Proposed a physics-aware interpretable machine learning framework for damage screening and fragility estimation.
Utilized data from the 2015 Gorkha earthquake, incorporating building inventory parameters and macroseismic intensity.
Employed gradient-boosted trees and split-conformal screening for evaluations.
Gradient-boosted trees achieved a balanced accuracy of 0.430 and a quadratic weighted kappa of 0.633.
Spatial transferability resulted in a balanced accuracy of 0.358 and a kappa of 0.471.
Split-conformal screening exhibited empirical coverage of 0.846, indicating reliable seismic triage capability.

Abstract

Rapid post-earthquake damage assessment is crucial for structural inspection prioritization and emergency response. This study proposes a physics-aware interpretable machine learning framework for seismic damage screening and fragility estimation using data from the 2015 Gorkha earthquake, Nepal. The framework integrates building inventory parameters, capacity-surrogate features, macroseismic intensity, and uncertainty quantification. Gradient-boosted trees achieved the best performance with balanced accuracy of 0.430 and quadratic weighted kappa of 0.633 under in-distribution validation, while spatial transferability produced balanced accuracy of 0.358 and kappa of 0.471. Split-conformal screening achieved empirical coverage of 0.846, demonstrating reliable portfolio-scale seismic triage capability in data-scarce regions.

KI fragen

Bookmark