Assessing the Loss of Building Functionality Post‐Earthquake Using Machine Learning

Assessing the loss of building functionality following a major earthquake event is crucial for establishing the post‐earthquake habitability of affected structures. Recognizing the stochastic uncertainties and interdependencies of component‐ and subsystem‐level damage on overall functionality, this study integrates extensive empirical damage data from the 2010–2011 Canterbury Earthquake Sequence (CES) in Christchurch, New Zealand, with advanced machine learning techniques to develop a novel computational framework for evaluating earthquake‐induced functionality loss. Central to this framework is the Random Forest (RF) model that systematically identifies critical components whose impairment significantly degrades building functionality and delineates the cascading damage pathways that culminate in discrete functionality state (FS) outcomes. By propagating both epistemic and aleatory uncertainties, the framework yields probabilistic FS estimates at individual‐building and regional scales, thereby underpinning agile, evidence‐based recovery planning. Validation through case studies confirms the framework's efficacy, revealing that damage to nonstructural components, such as parapets, ornamentation, cladding, and glazing, exerts disproportionately large effects on building functionality. Moreover, five principal damage‐propagation pathways have been elucidated, each mapping a unique sequence of component failures to a specific FS, thus illuminating the complex interdependencies that govern post‐earthquake usability. The framework and its associated findings provide insights into the recovery‐based design and can be used by engineers in rapid damage assessments to inform re‐occupancy or repair decisions of local authorities and building owners.