1. Introduction
In the last few years, the application of Machine Learning (ML) technologies has been widely investigated in several fields of optical communications and networking, including e.g. lightpath Quality of Transmission (QoT) estimation, automated fault management, routing and wavelength/spectrum assignment [1]. However, the vast majority of ML approaches proposed in the scientific literature are used as “black boxes”, i.e., they do not expose their internal mechanics nor the decisional processes adopted to associate the produced outputs with the set of feature values provided as input. This hinders the interpretability of the models and prevents the extraction of useful insights that could be leveraged to better understand the nature of the problem at hand.