ABSTRACT
The recovery time (RT) is one of the essential components of infrastructure seismic resilience analysis. This seismic infrastructure resilience is crucial to keep the functionality of critical infrastructure and recover them quickly during strong earthquakes. It is necessary to ensure that the RT is estimated beforehand. This RT depends on some factors (Derras & Makhoul 2022). We develop machine-learning empirical models for estimating the RT based on earthquake damage data to achieve that goal. Firstly, a dataset of earthquake damage was established. This dataset contains the metadata (explanatory factors), such as earthquake magnitude (M), Ground-Motion Intensity Measures (IM), as well as the recovery time (target). Using the more famous machine learning algorithms, we establish empirical models to predict the RT of a given system (gas, power, water, and Telecommunications). The best model must represent well the resilience curve with an optimal aleatory variability. The estimate early of the recovery time (here by machine learning approach) is essential information for authorities and a primary step in the infrastructure seismic resilience assessment.
