Towards an Adaptive Education through a Machine Learning Recommendation System

Educational institutions have a tremendous burden of handling students with low academic performance (At-risk students). Many approaches support this group of pupils, such as psychological therapy, a proper timetable for vulnerable pupils, recall, personal training, mock-tests, direct private education, or success centres. However, these methods are not enough to solve the issue since other factors influence the learner’s success, which could be their family difficulties, cognitive style, prior performance, and college foundation level. This paper explores machine learning models for predicting at-risk students and then build a recommendation platform to provide direct system-based coaching to such students to remediate the fragmentation in their knowledge and skills. We use a dataset of 554 students from a computer program; the study aims to break down the curricula into a set of crumbs of knowledge and skills used to measure learners’ progress during their study. We used various machine learning algorithms, a decision tree with an accuracy of 82.68% (positive class: Good Standing), and high coverage of 87.69% (positive class: Good Standing). We completed the model optimization and used the ROC comparison to compare the classifier models. A remediation algorithm is used to support at-risk cases, leading to a sharp decline in the at-risk rate. The study finds that the current applied approaches to handle at-risk students exaggerate the problem, and students should not be treated in bulk.