Convolutional Neural Network and GridSearch for Predicting Learner Performance in E-Learning

Abstract

Predicting student performance earlier is among the most significant research areas in e-learning. Accurate prediction allows educators and stakeholders to take the appropriate precautions and make decisions ahead of time to enhance learning outcomes. Numerous previous studies have employed conventional machine learning processes. This work investigated the effectiveness of a convolutional neural network (CNN) to predict learner performance in a virtual learning environment (VLE). This STUDY employed three independent feature selection methods: 1) CNN, 2) random forest (RF), and 3) autoencoder (AE). The GridSearch cross validation (GridSearchCV) strategy explored crucial model hyperparameters that optimize performance. This paper uses the Open University learning analytics dataset (OULAD), which contains data from 32,593 learners enrolled in 22 courses offered by the Open University. The synthetic minority oversampling technique (SMOTE) resolved the issue of an imbalanced dataset. Stratified cross-validation splits the data into training, validation, and testing sets. The proposed approach reached an accuracy of 88% in multi-class classification and 94% in binary classification, outperforming all previous studies that utilized Open University learning analytics dataset.