Behavioral Feature Analysis For Learner Affect Identification

The paper presents methodology for mapping of learner interaction patterns in virtual learning system to the affective state. A learning activity can lead the learner to confused or confident state of mind. This affective state of learner can be identified by interaction pattern analysis. The examination mode of learning system is designed with questionnaire based on Bloom's taxonomy cognitive levels. We present an approach for establishing relationship between cognitive level test performance of learner and affective state. Various non-intrusive interaction parameters captured during learning activity act as input features. We find that random forest algorithm provides very good accuracy in determining affective state.