Evaluating Machine Learning Techniques for Improved Adaptive Pedagogy

Literature has shown that learning gains may be improved significantly if students are offered individual attention. The traditional offering of such individualised attention is however not always possible due to limited resources that lead to high student-to-teacher ratios. One solution to addressing this gap in education has been the proposal of Intelligent Tutoring Systems which promised to deliver adaptive learning experience to improve student learning outcomes. Unfortunately, results have been less than satisfactory - one factor attributing to this poor performance is that such systems fail to adapt to varying student learning needs. To investigate how this poor performance can be improved upon, this paper examines several incremental machine learning techniques by incorporating them in intelligent tutor systems to evaluate their performance for adaptive pedagogy to determine the most appropriate incremental machine learning technique for implementation in intelligent tutor systems. Support Vector Machines (SVMs), k-Nearest Neighbour (k-NN) and the Naïve Bayes algorithm where investigated and put into practical tests. Results obtained from training the three methods using the same datasets revealed that the naïve Bayes outperformed the k-NN and the SVMs. The results led to the conclusion that Naïve Bayes would perform well when implemented for pedagogical decision-making in ITSs. It was therefore recommended that developers use effective incremental machine learning algorithms such as the naïve Bayes, if intending to build adaptive ITS that can be used by African Students.