A survival analysis based volatility and sparsity modeling network for student dropout prediction

Area of Science:

• Machine Learning
• Educational Data Mining
• Survival Analysis

Background:

• Student Dropout Prediction (SDP) is crucial for Massive Open Online Courses (MOOCs).
• Traditional binary classification models offer limited temporal insights.
• Existing survival analysis methods struggle with data volatility and sparsity, impacting performance.

Purpose of the Study:

• To propose an end-to-end deep learning framework, SAVSNet, for robust Student Dropout Prediction.
• To address data volatility and sparsity challenges inherent in MOOC data.
• To improve the accuracy and consistency of dropout predictions over time.

Main Methods:

• Developed a Survival Analysis based Volatility and Sparsity Modeling Network (SAVSNet).
• Employed Convolutional Neural Networks (CNNs) for time series smoothing and Long-Short Term Memory (LSTM) networks for data preservation.
• Introduced a Time-Missing-Aware LSTM unit to effectively handle data sparsity by incorporating missingness patterns.
• Utilized a survival analysis loss function for parameter estimation, ensuring monotonically decreasing survival probability outputs.

Main Results:

• SAVSNet demonstrated superior performance compared to state-of-the-art methods on two real-world MOOC datasets.
• The model effectively smoothed volatile time series data while preserving essential information.
• The Time-Missing-Aware LSTM unit successfully mitigated the negative impact of data sparsity.
• The proposed framework achieved more accurate and consistent dropout predictions.

Conclusions:

• SAVSNet offers an effective end-to-end deep learning solution for Student Dropout Prediction in MOOCs.
• The model's ability to handle data volatility and sparsity significantly enhances prediction accuracy.
• This approach provides a valuable tool for proactive intervention and student support in online learning environments.