We describe how to use Machine Learning techniques in a digital game-based system for student assessment. The students attend training courses to apply for a professional certification. A digital game-based system is designed to evaluate whether a student has the necessary knowledge for the certification.  The students are evaluated as expert players or non-expert players according to their performance. We focus on how to use Machine Learning techniques in the process of automation of the evaluation of game-playing episodes. We propose two possible solutions for this project: Hidden Markov Model and association rule mining. The goal of this project is to provide an industrial-grade solution, capable of assessing hundreds of students per day, in a fully automatic mode.