<![CDATA[Accurate classification of building damage following disasters plays a critical role in facilitating efficient rehabilitation and reconstruction. Traditional field-based assessment methods, however, present significant limitations—including time inefficiencies, susceptibility to subjective interpretation, and potential safety risks for survey personnel. Recent advancements in machine learning (ML) have significantly improved the efficiency and objectivity of post-disaster damage assessment by leveraging diverse data sources such as satellite imagery, unmanned aerial vehicles (UAVs), and even crowdsourced social media content. This study conducts a narrative literature review of 78 peer-reviewed articles published from 2020 to 2024, focusing on ML-driven methodologies for classifying building damage and generating rehabilitation recommendations. The literature review reveals a prevailing reliance on deep learning models—especially convolutional neural networks (CNNs) and transformer-based architectures—due to their robust accuracy and adaptability across varied disaster scenarios. Furthermore, novel approaches like self-supervised learning, ensemble methods, and few-shot learning show promising potential in addressing challenges posed by sparse or unevenly distributed datasets. Despite rapid advancements in ML-based post-disaster building damage classification, real-world implementation remains constrained. This review synthesizes current trends, persistent challenges, and critical research gaps to inform the development of a robust ML framework for post-disaster recovery efforts. This study uniquely highlights the integration of ML-based classification with rehabilitation planning frameworks, providing practical guidance for disaster management agencies to optimize post-disaster recovery strategies.]]>
Copyrights © 2026
