CNN autoencoders and LSTM-based reduced order model for student dropout prediction
- Publisher:
- Springer Nature
- Publication Type:
- Journal Article
- Citation:
- Neural Computing and Applications, 2023, pp. 1-17
- Issue Date:
- 2023-01-01
Embargoed
| Filename | Description | Size | |||
|---|---|---|---|---|---|
| CNN autoencoders and LSTM-based reduced order model for student dropout prediction.pdf | Accepted version | 3.26 MB |
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is currently unavailable due to the publisher's embargo.
In recent years, Massive Open Online Courses (MOOCs) have become the main online learning method for students all over the world, but their development has been affected by the high dropout rate for a long time. Therefore, dropout prediction is a vital task for early teaching intervention and user retention. The students’ learning records are stored in MOOCs, which contain high-dimensional time series features. However, these features are hard to process, and the nonlinear relationship between the features is difficult to learn. These limitations have become obstacles to improve the performance in dropout prediction. In this paper, we propose a new neural dimension-reduced dropout prediction model based on neural network model to address the limitations. The proposed model, called CNNAE-LSTM, is constructed by convolutional neural network autoencoder (CNNAE) and long short-term memory neural network (LSTM). Specifically, CNNAE-LSTM compresses the students’ learning features into a low-dimensional latent space for reconstruction through CNNAE, then projects the latent space, retains the representative features in the learning records, and finally minimizes the reconstruction error to obtain the nonlinear relationship between features and dropout. The introduced LSTM neural network can obtain the time evolution of its latent vector. Our experiments on the KDD CUP 2015 dataset and the real-world dataset XuetangX demonstrate that the proposed model exhibits better predictive performance compared to the state-of-the-art baseline methods.
Please use this identifier to cite or link to this item: