Expected goal models (xG) are of great importance as they are the most accurate predictor of future performance of teams and players in the world of soccer. This metric can be modeled by machine learning, and the models developed consider an increasing number of attributes, which increases the cost of learning it. The use of xG is not widespread in handball, so the question of learning it for this sport arose, in particular which attributes are relevant for learning. Here, we used a wrapper approach to determine these relevant attributes and guide teams through the data collection stage.