A Comparative Evaluation of Adaptive Learning Recommendation Models for Improved Student Learning Performance

Adaptive learning models utilise interaction data from learners to recommend study resources, practice activities and learning pathways. This paper provides a comparative evaluation of the three main categories of models; knowledge tracing, reinforcement learning and language model assisted. These models were evaluated in terms of how they conceptualise knowledge, choose the next activity, provide explanations for recommendations and overcome cold start and fairness issues. A standardised experiment protocol was developed and includes data processing, time based split of training, validation and testing data sets, leakage control, prediction accuracy and ranking metrics and learning performance metrics. As there was no original dataset, implementation, sample of participants or experiment carried out for this manuscript, the results section details the validation plan. The aim is to give an open and replicable design for evaluating adaptive recommendation models. This paper also outlines the requirements for keeping track of assumptions, model choices, learner feedback and ethical considerations for implementation in an educational setting.