Predicting Learning Performance in Serious Games

The prediction of learning performance is an important task in the context of smart tutoring systems. A growing community from the field of Learning Analytics and Educational Data Mining investigates the methods and technologies to make predictions about the competencies and skills, learners may reach within a specific course or program. Such performance predictions may also enrich the capabilities and the effectiveness of serious games. In game-based assessment, predictions add a novel dimension for the personalization and adaption in games for which these functions may provide a valuable data basis. The Learning Performance Vector (LPV) allows utilizing information about the learning domain (i.e., the competencies and the structure of competencies) and log file information from games to make performance predictions. In a simulative study based on existing datasets, we explored the characteristics of the approach and compared it to a linear regression model. The results indicate that the LPV is a promising method, specifically in data rich game-based scenarios with limited external information.