Student Modeling in Real-Time during Self-Assessment Using Stream Mining Techniques

In order to personalize the assessment services, the assessment systems need to build suitable student models for heterogeneous student populations. The present study focuses on efficiently modeling students according to their time-varying behavior during web-based self-assessment, enriching the models with a notion of dynamics. The suggested approach forms and revises the student models on-the-fly, using three popular stream mining classification techniques. All methods use specific time-based features as predictors, and the students' self-assessment achievement levels as target values. The obtained results demonstrate that level of certainty, effort and time-spent on answering correctly/wrongly could contribute to pursuing fine-grained and robust student models during self-assessment.