Educational games are addressed to a large variety of learners who differ, for example, in their playing style preferences, their learning styles, and their personalities. Accordingly, the learning-playing process is different from a learner to another. Therefore, an insistent need is born for modeling the learners’ behaviors. Taking into account such differences can enhance the learning process, the motivation and the engagement of the learner in educational games and thus a better convergence to the learning achievements. This paper presents a preliminary work towards adaptation in educational games based on a learner’s play type. Modeling learners’ types by using an explicit method, namely questionnaire, can make them not motivated especially when questionnaires are typically too long. In this context, we proposed a new implicit approach based on data mining which models the learner’s type while learning-playing based on his/her generated traces in an Educational Massively Multiplayer Online Role Playing Game (EMMORPG). In particular, during the learning-playing process, the learners' generated data were collected and then pre-processed. After that, we applied a decision tree classifier on the prepared data in order to build our prediction model. A study was carried out to validate the proposed approach, with sixty learners (40 females and 20 males) aged between 19 and 22. They used a new developed educational game called “Educational Browser Quest” to learn the basics of web programing. The obtained results showed that the performance measures of our model in predicting the learners’ type are above 70%.