This study proposes a machine-learning-assisted approach for developing predictive models of the building usability class and observing reconstruction costs following seismic events. The training data used here were derived from surveys conducted after the 2009 L'Aquila earthquake, collected via Rapid Post-Earthquake Damage Evaluation (AeDES) forms. These forms document key structural and nonstructural characteristics, along with observed damage, across seven sections, resulting in approximately 60 categorical features. Additionally, seismic intensity measures are incorporated to enhance prediction accuracy, providing physics-based insights into the structural demands experienced by the surveyed buildings. The machine learning models are trained to predict the post-earthquake usability classification of buildings across four classes: immediately accessible (class A), usable only after emergency interventions (class B), partially unusable (class C), and fully unusable (class E). The robustness and interpretability of the resulting ML models are also assessed, considering also the reconstruction costs derived by the simplified vulnerability index assessment. Findings from this study underscore the potential of machine learning techniques to support rapid post-emergency response efforts as well as to facilitate reliable simulations at a regional scale, contributing to cost-effective planning and seismic risk mitigation strategies.