MASONRY BUILDINGS RECONSTRUCTION COST POST-EARTHQUAKE ANALYSIS WITH MACHINE LEARNING

Machine learning (ML) methods represent advanced probabilistic data-driven techniques that can potentially empower earthquake engineering-related practices thanks to their ability to analyze hidden patterns buried in data. In this study, the authors examined the adoption of ML classification models for predicting reconstruction costs observed after L’Aquila 2009 earthquake event in the Abruzzi region based on masonry building vulnerability indexes and usability classes. The latter data mainly refers to the Rapid Post-Earthquake Damage Evaluation (AeDES) forms surveys, containing about 60 categorical features. Additionally, seismic intensity measures (IMs) have been incorporated into the database to encompass physics-based data characterizing the input demand. After an initial exploratory data analysis, the authors conducted some preliminary analysis on using the sole vulnerability index data for predictive purposes of the reconstruction costs using random forest algorithms. This approach could potentially benefit public administrations for long-term optimal resource earmarking on a regional scale. Indeed, preliminary findings from this study underscore the potential of using those MLassisted procedures for quick simulation scenarios based on various reconstruction cost classes, contributing to cost-effective planning and seismic risk mitigation strategies. However, in this preliminary work, the AeDES usability data have not been considered due to the purely predictive purpose conjecture carried on in this study. Therefore, future promising developments should also combine the vulnerability index information and the observed building usability classes, to improve the extrapolation performance of ML-assisted predictive tools