Abstract. The present work proposes and validates a method to infer the learning zone of CS1 students in an online judge. To do so, the student's grade will be predicted, using a programming profile based on the data left by them as they solve exercises in that system. Students who scored lower than 5 were classified in a difficulty zone, otherwise in an expertise zone. Machine learning algorithms were used to make the prediction. The proposed predictive model obtained an accuracy of 78.3 % at the first two weeks of class, which overcomes research results that were conducted in similar scenarios.