Earthquake damage assessment in Nepal using machine learning and SHAP: a regional analysis of building vulnerabilities

Source
DataCite
Published
2026
Country
Nepal
License
Creative Commons Attribution 4.0 International
Themes
Natural Disasters, Energy Infrastructure, Disaster Risk & Resilience

Get the data from DataCite

About this dataset

Earthquake damage assessment is essential for effective disaster response and resilience planning. Many existing machine learning models lack the engineering interpretability necessary for practical decision-making. This research presents an interpretable framework applied to the 2015 Gorkha earthquake in Nepal, employing XGBoost and SHAP (Shapley Additive exPlanations) to analyse data from over 760,000 buildings across 11 districts. The model predicts multi-class damage states based on structural and geometric attributes, including age, height, plinth area, and types of roofs, foundations, floors, and superstructures. The results indicate that XGBoost achieves high sensitivity in identifying severe damage, which supports prioritisation of collapse risks. SHAP analysis highlights key factors such as mud-mortar stone superstructures, age, height, and plan irregularity as primary...

How the data was collected

The source metadata does not specify a detailed collection methodology. Provenance: this record was harvested from DataCite and published by Taylor & Francis. Consult the original source for methodological details, temporal coverage, and access constraints.

Browse sustainability datasets on SusanHub