Game Learning Analytics can be used to conduct evidence-based evaluations of the effect that serious games produce on their players by combining in-game user interactions and traditional evaluation methods. We illustrate this approach with a case-study where we conduct an evidence-based evaluation of a serious game's effectiveness to increase awareness of bullying. In this paper, we describe: (1) the full process of tracking in-game interactions, analyzing the traces collected using the standard xAPI-SG format, and deriving game learning analytics variables (to be used as "evidences"); and (2) the use of those variables to predict the increase in bullying awareness. We consider that this process can be generalized and replicated to systematize, and therefore simplify, evidence-based evaluations for other serious games based on the interaction data of their players.