Various didactic strategies to develop Computational Thinking (CT) skills have been successful in terms of student engagement and educational outcomes. However, monitoring the learning progress of students is still a hurdle to teachers and researchers. In this context, we explore the use of self-organizing maps for analyzing games produced in a game building workshop offered simultaneously to technical education students in Brazil and to undergraduate students in Computer Engineering in Chile. Metrics from seven CT features present in the games were extracted with the Dr. Scratch tool and used as an input in the training process. The results allowed a clustering analysis considering the identified features and the correlation between learning behaviors. The organization of the map reflected a progressive skill acquisition identified by features present in the developed games. Also, it could be identified that students of both educational levels reached similar levels of CT skill development.