Do you remember the ESA Collision Avoidance Challenge last year? HRI-EU made 1st place by applying a variety of machine learning methods and data engineering for predicting the collision risk between satellites and space debris in low earth orbit.
Since then, a post-competition analysis by the organizers and top performing teams resulted in further insights. Their findings on the general feasibility of machine learning methods for this application and future competition improvements are now published as a paper.
Take a look here.