To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Temporal-Based Transformed Recurrent Neural Network for Dropout Prediction in Online Learning |
|
Author |
Abdullah Alshehri |
| Citation |
Vol. 22 No. 9 pp. 772-783 |
|
Abstract |
Online learning has provided flexible learning opportunities worldwide to a wide range of individuals. The rapid advances in the online learning industry have allowed numerous universities and institutes to offer online courses that provide students with adequate learning experiences. Although online learning provides great advantages to individuals, the increase in dropout and course incompletion rates brings a crucial challenge that adversely influences the effectiveness of online learning. To address the dropout issue, many studies have been directed at student performance prediction to determine the intention to dropout. The prediction of student performance relies on the analysis of log data learning. However, the sparsity and high dimensionality of learning data have brought a complex feature extraction to build a reliable prediction model. Moreover, the learning behavior may change over time due to the change in learning settings for various reasons, such as using different devices to complete the course. Nevertheless, the current prediction models fail to capture the subtle change in learning behavior where the prediction can significantly be improved. This paper presents a novel model to develop a prediction model for student dropout in online learning. The model proposes a novel Transformed Remember Gated-based Long Short-Term Memory TRG-LSTM to structure temporal feature space from multivariate time series learning activities. TRG-LSTM is employed to map the log data of learning activities in non-linear temporal domain as multi-sequences representation. Thus, the behavior learning profile is built upon non-linear feature space thus to learn the inter-relations between temporal dependencies. Moreover, the proposed TRG-LSTM can handle the subtle changes in learning behavior by modifying the activation function value of remember gate to capture subtle information over temporal granularity. The evaluation, using online learning dataset, has shown that the proposed model has outperformed the benchmarked models to predict student dropout more accurately. |
|
Keywords |
Deep Learning, LSTM, Online Learning, Student Performance Prediction. |
|
URL |
