Abstract

This research presents a machine-learning-based framework for classifying earthquake-induced structural damage, addressing issues like class imbalance, noisy data, and lack of labelled samples. Ensemble classifiers like Random Forest, XGBoost, and Multi-Layer Perceptron are used to increase classification accuracy. Shapley Additive explanations (SHAP) are used to identify key damage levels. The model's effectiveness in emergency decision-making is evaluated using actual post-earthquake construction data from Nepal. The method provides evidence-based assistance for focused treatments, focusing on building age, material, anomalies, and soil type. The proposed structural evaluation process enhances conventional methods and offers an automated, scalable way to conduct accurate and quick structural assessments in disaster-risk locations. The focus on explainability of models ensures that outcomes are trustworthy and actionable for stakeholders in disaster management and urban planning agencies.