Evaluating Deep Sequential Knowledge Tracing Models for Predicting Student Performance

Abstract

One of the key priorities in all instructional environments is to ensure that students recognise their learning mechanisms and pathways. Knowledge Tracing (KT), the task of modelling student knowledge from their learning history, is an important problem in the field of Artificial Intelligence in Education (AIEd) and has numerous applications in the development of interactive and adaptive learning technologies. KT can be utilised to understand each student’s distinct learning style, particular needs, and ability levels. We trained and evaluated the performance of Knowledge Tracing models on the ASSISTments dataset and EdNet-KT1 dataset. This study revealed that deep learning models for knowledge tracing (Deep Knowledge Tracing (DKT), Dynamic Key-Value Memory Network (DKVMN), and Attentive Knowledge Tracing (AKT)) outperform the Markov process model (Bayesian Knowledge Tracing). We also observed that AKT and DKT go hand in hand with predicting whether or not the following question will be answered correctly or incorrectly by the student.