Search for a command to run...
Serious games have become an important tool to train individuals in a range of different skills. Importantly, serious games or gamified scenarios allow for simulating realistic time-critical situations to train and also assess individual performance. In this context, determining the user's cognitive load during (game-based) training seems crucial for predicting performance and potential adaptation of the training environment to improve training effectiveness. Therefore, it is important to identify in-game metrics sensitive to users' cognitive load. According to Barrouillets' time-based resource-sharing model, particularly relevant for measuring cognitive load in time-critical situations, cognitive load does not depend solely on the complexity of actions but also on temporal aspects of a given task. In this study, we applied this idea to the context of a serious game by proposing in-game metrics for workload prediction that reflect a relation between the time during which participants' attention is captured and the total time available for the task at hand. We used an emergency simulation serious game requiring management of time-critical situations. Forty-seven participants completed the emergency simulation and rated their workload using the NASA-TLX questionnaire. Results indicated that the proposed in-game metrics yielded significant associations both with subjective workload measures as well as with gaming performance. Moreover, we observed that a prediction model based solely on data from the first minutes of the gameplay predicted overall gaming performance with a classification accuracy significantly above chance level and not significantly different from a model based on subjective workload ratings. These results imply that in-game metrics may qualify for a real-time adaptation of a game-based learning environment.