An Enhanced CNN Model on Temporal Educational Data for Program-Level Student Classification

Abstract

In educational data mining, study performance prediction is one of the most popular tasks to forecast final study status of students. Via these predictions, in-trouble students can be identified and supported appropriately. In the existing works, this task has been considered in many various contexts at both course and program levels with different learning approaches. However, the real-world characteristics of the task’s inputs and output such as temporal aspects, data imbalance, and data shortage with sparseness have not yet been fully investigated. Making the most of deep learning, our work is the first one handling those challenges for the program-level student classification task on temporal educational data. In a simple but effective manner, a novel solution is proposed with convolutional neural networks (CNNs) to exploit their well-known advantages on images for temporal educational data. Moreover, image augmentation is done in different ways so that data shortage with sparseness can be overcome. In addition, we adapt new loss functions (Mean False Error and Mean Squared False Error) to make CNN models tackle data imbalance better. As a result, the task is resolved by our enhanced CNN models with more effectiveness and practicability. Indeed, in an empirical study on three real temporal educational datasets, our models outperform other traditional models and original CNN variants on a consistent basis with Accuracy of about 85%–95%.