Detecting and Modelling Stress Levels in E-Learning Environment Users
A modern Intelligent Tutoring System (ITS) should be sentient of a learner's cognitive and affective states, as a learner’s performance could be affected by motivational and emotional factors. It is important to design a method that supports low-cost, task-independent and unobtrusive sensing of a learner’s cognitive and affective states, to improve a learner's experience in e-learning, as well as to enable personalized learning. Although tremendous related affective computing research were done in this area, there is a lack of empirical research that can automatically measure a learner's stress using objective methods. This research is set to examine how an objective stress measurement model can be developed, to compute a learner’s cognitive and emotional stress automatically using mouse and keystroke dynamics. To ensure the measurement is not affected even if the user switches between tasks, three preliminary research experiments were carried out based on three common tasks during e-learning − search, assessment and typing. A stress measurement model was then built using the datasets collected from the experiments. Three stress classifiers were tested, namely certainty factors, feedforward back-propagation neural network and adaptive neuro-fuzzy inference system. The best classifier was then integrated into the ITS stress inference engine, which is designed to decide necessary adaptation, and to provide analytical information of learners' performances, which include stress levels and learners’ behaviours when answering questions.
- PhD