Enlarging the sample and sky coverage of AGN observations with reliably estimated physical parameters is particularly important for multimessenger astronomy, where signals from individual sources are often weaker, such that searching for correlations between a population class (e.g, AGN) and a messenger (e.g., neutrinos or cosmic rays) is common. However, knowledge of physical parameters of AGN, such as the mass of the central black hole M$_{\rm BH}$ and the Eddington ratio $\lambda_{\rm Edd}$, are limited by the feasibility of large spectroscopic follow-up surveys. We show an application of machine learning (ML) techniques to reconstruct AGN physical parameters using multi-wavelength photometric observations only, in the soft X-ray, mid-infrared, and optical bands, as a way to increase the number of characterized AGN. We present a catalogue of 21 364 newly reconstructed AGN, with redshifts ranging from 0 < z < 2.5. The redshift $z$, the bolometric luminosity L$_{\rm Bol}$, M$_{\rm BH}$, $\lambda_{\rm Edd}$, and AGN class (obscured or unobscured) are estimated with their associated uncertainty, using a Support Vector Regression and Random Forest algorithms.