Which user interactions predict levels of expertise in work-integrated learning?

Predicting knowledge levels from user’s implicit interactions with an adaptive system is a difficult task, particularly in learning systems that are used in the context of daily work tasks. We have collected interactions of six persons working with the adaptive work-integrated learning system APOSDLE over a period of two months to find out whether naturally occurring interactions with the system can be used to predict their level of expertise. One set of interactions is based on the tasks they performed, the other on a number of additional Knowledge Indicating Events (KIE). We find that the addition of KIE significantly improves the prediction as compared to using tasks only. Both approaches are superior to a model that uses only the frequencies of events.