AHP-Guided Stacked Ensemble Modeling for Student Engagement Level Prediction in Online Education

Given the increasing importance of online education, this research seeks to provide a solution to more accurately predict the level of student engagement in these environments. The main goal is to improve the performance of prediction models using an innovative hybrid approach. The proposed method consists of three key steps: first, data preprocessing to prepare and refine a set of indicators related to the level of student engagement in online education environments. Second, the use of a novel feature selection method based on Analytic Hierarchy Process (AHP) and an iterative approach. In this step, AHP is used to rank and weight the features, and then an iterative process is used to determine an appropriate number of these ranked features. This is done in order to select the most effective features for prediction. Third, the use of an ensemble stack based on classical machine learning models, including K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Naive Bayes (NB). In this stack, the learning models are organized at two levels. At the first level, KNN, SVM, and NB models are used independently to predict the target variable (engagement level) based on the selected features. At the second level, a Multilayer Perceptron (MLP) model as a meta-learner makes the final prediction based on the results of the predictions made by the first-level models. This combined structure, in order to take advantage of the strengths of each model and reduce errors, ultimately leads to more accurate and efficient prediction of student engagement level in online learning environments. According to the proposed method, experiments were conducted on a real dataset and the results obtained indicate the very good performance of the presented model. The values obtained for both Precision and Recall are above 90%. These results show that the proposed model is able to predict different levels of student engagement with high accuracy and, at the same time, has an acceptable performance in terms of both Precision of predictions and Recall of complete samples.