An ever-growing number of applications employ machine learning models for prediction, decision-making, or state estimation. This thesis proposes a specific certification process for machine learning-based systems performing regression tasks, in order to support future applications. The analysis is specially targeted to safety-critical systems, such as those involved in aviation, with the aim of showing compliance with the latest EASA safety requirements in regard to Artificial Intelligence (AI). The intent consists in providing a general framework that can be adapted to different contexts and, when possible, referred to the standard processes for traditional components, i.e., not AI-based, described in the EUROCAE/SAE guidelines ED-135/ARP4761A and ED-79B/ARP4754B.