Multi-label classification and explanation methods for students’ learning style prediction and interpretation

Abstract

The current paper attempts to describe the methodology guiding researchers on how to use a combination of machine learning methods and cognitive-behavioral approaches to realize the automatic prediction of a learner’s preferences for the various types of learning objects and learning activities that may be offered in an adaptive learning environment. Generative as well as discriminative machine learning methods may be applied to the classification of students’ learning styles, based on the student’s historical activities in the e-learning process. This paper focuses on the discriminative models that try to learn which input activities of the student(s) will correlate with a particular learning style, discriminating among the inputs. This paper also investigates several interpretability approaches that may be applicable for the multi-label models trained on non-correlated and partially correlated data. The investigated methods and approaches are combined in a consistent procedure that can be used in practical learning personalization.