Predicting Student Performance in Experiential Education.

Experiential learning is a key development area of artificial intelligence in education (AIEd). It aims to provide learners with intuitive environments for autonomous knowledge formation and discovery through interactive experiences. However, experiential learning in AIEd faces two main challenges. Firstly, measuring learning performances in unstructured and informal educational settings is difficult. Secondly, providing frequent or timely feedback on student performance is inefficient. To address these issues, this paper explores using natural language processing (NLP) and the tool for the automatic analysis of cohesion (TAACO) features as indicators of student performance in an experiential learning course. Both NLP and TAACO features were tested on a baseline CART decision tree (DT) machine learning (ML) model with and without a grade population distribution mask to predict student final scores at the end of the course. Our results show that (1), the use of a distribution specific Gaussian mask significantly increases prediction accuracy of the CART DT. (2), NLP and TAACO features provide high information value for ML prediction tasks. (3), the CART DT is able to accurately classify learner grade scores against human assessments.