ANN-LSTM: A deep learning model for early student performance prediction in MOOC.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Data Science

Background:

• Learning analytics seeks to predict student performance for timely interventions, but challenges exist in virtual environments due to distance.
• Existing predictive models for Massive Open Online Courses (MOOCs) often use binary classification and provide results late in the course, hindering timely interventions.

Purpose of the Study:

• To develop and validate a day-wise, multi-class model for predicting student performance in MOOCs.
• To address the limitations of existing models by enabling earlier and more granular performance predictions.

Main Methods:

• Proposed a novel Artificial Neural Network and Long Short-Term Memory (ANN-LSTM) model for day-wise, multi-class student performance prediction.
• Compared the ANN-LSTM model against baseline models: Recurrent Neural Network (RNN) and Gated Recurrent Unit (GRU).
• Evaluated ANN-LSTM performance against state-of-the-art models using accuracy metrics.

Main Results:

• The ANN-LSTM model achieved the highest accuracy among the baseline models, reaching approximately 70% by the third month.
• ANN-LSTM outperformed RNN (53% accuracy) and GRU (57% accuracy) models significantly.
• ANN-LSTM demonstrated enhanced accuracy rates of 6-14% compared to existing state-of-the-art models.

Conclusions:

• The ANN-LSTM model effectively predicts student performance in MOOCs, offering early insights through its time-series capabilities.
• The model's architecture, particularly LSTM's ability to retain latent dependencies, is crucial for accurate, early performance predictions.
• This approach facilitates more timely and effective instructor interventions in virtual learning environments.