Tibia Player Profiling and Modeling using Machine Learning

User engagement is one of the critical factors for the success of software systems. The more interactive a system is, the higher the chance of achieving its goals. Recently, the number of users who play Multiplayer Online Role-Playing Games (MMORPG) has become massive. This growth in user numbers led to different player types with different preferences and skills. Automatic modeling of player types enables the design of adaptive games and consequently improves the player experience. This paper proposes a framework for profiling and modeling the player's behavior. In-game players' data were collected from the Tibia game open server. Player profiles were built with the assistance of Tibia experts. Then six classifiers Random Forests, Artificial neural networks (ANNs), Stochastic gradient descent (SGD), Decision trees, Naïve Bayes, and K-nearest neighbors (KNN), were trained using the built profiles to model the player types. Finally, the Bartle test was given to sample players; the profiles of these players were used to test the previously trained player-type classifiers. Experimental results showed that a classification accuracy equal to 90% could be achieved.