Machine Learning Framework for Performance Analysis of Students Using Content Recommendation in Education System

Abstract

As digital learning environments continue to expand there is a vast amount of information about student interactions – and significant opportunities and challenges for personalised learning. In this paper, a novel framework is presented, called ML-PACS (Machine Learning-based Performance Analysis and Content Suggestion), which comprises an ensemble learning, deep learning and a hybrid collaborative-content-based filtering to simultaneously predict student academic performance and suggest content for students in a context-aware way. The framework receives data streams which are heterogeneous like previous academic records, Logs from Learning Management System (LMS), Quiz and Assignment scores, Learning style indicators, Peer interaction metric. The stacked ensemble of three classifiers: XGBoost, Random Forest and Long Short-Term Memory (LSTM) networks obtained an accuracy of 92.8%, an F1 score of 91.2%, a precision of 91.6%, and a recall of 90.9% with 2400 learners across six semesters. The best Hit Rate@10 (83.9%) and NDCG@10 (0.567) among all the baseline methods are achieved by the recommendation module with a performance-aware hybrid filter with the prediction of the performance tier as a contextual signal. The results of the longitudinal evaluation reveal that there is a different average score for the experimental and control groups for the three academic years evaluated, showing an improvement of 80.1% in the experimental group and 62.0% in the control group. The suggested model provides specific recommendations to the teachers who have automated alert systems and visual dashboards, for early pedagogical interventions. Results reveal that, when predictive analytics is used alongside adaptive content recommendation, students' outcomes measurably improve in the long run.