Predicting Students At-Risk Using Deep Learning Neural Network: a comparison of performance of different models

Abstract

The availability of educational data, along with technology-enhanced learning platforms, allows for the extraction of students' learning behavior, mitigation of their worries, optimization of the educational environment, and data-driven decision-making. To detect at-risk students, this study uses a deep learning neural network trained on a set of attributes taken from a learner activity tracker application called experience API (xAPI). Other probabilistic models such as Random Forest, Support Vector Machine, and k Nearest Neighbours were compared to the convolutional neural network and multilayer perceptron models in correctly classifying at-risk students, and information gain ranking was used to eliminate features that are said to be redundant or irrelevant. After 10 fold cross-validation, the Support Vector Machine model outperformed all other models, including the Convolutional Neural Network and the Multilayer Perceptron, with a peak accuracy score of 77.51%, while the convolutional neural network and the multilayer perceptron achieved accuracy scores of 72.58% and 73.73%, respectively. Student behavioral features, according to the research, are useful predictors of student performance.