Damage classification after the 2009 L'Aquila earthquake using multinomial logistic regression and neural networks

Post-earthquake surveys represent a fundamental tool for managing the emergency phase after a strong earthquake. In Italy, the evaluation of the post-earthquake functionality of ordinary buildings is based on the AeDES forms (Agibilita e Danno nell'Emergenza Sismica, or equivalently, Rapid Post-Earthquake Damage evaluation forms). This form includes information on the building and records of the observed damage classified according to type and intensity in 60 subclasses. Based on the observed damage and expert judgment, the buildings are clustered into six risk classes, from A to F. The assigned class is used to calculate the maximum economic reimbursement owed for the reconstruction or repair of the building. However, often the cluster assignment is not entirely objective due to the inherent responsibility associated with a less conservative assessment. This paper uses the data from the 2009 L'Aquila earthquake to develop classification models based on multinomial logistic regression (MLR) and artificial neural networks (ANN) calibrated with data theoretically less influenced by personal biases. The proposed models, particularly the MLR, are intended to support the decision-making of the evaluation team in future updates of the AeDES forms. This approach cannot substitute expert evaluation, which is always necessary for complex scenarios but may mitigate the impact of subjectivity and can provide an indication of the expected outcome of the survey.