A Survival Analysis based Volatility and Sparsity Modeling Network for Student Dropout Prediction v1

Student Dropout Prediction (SDP) is of pivotal significance in mitigating withdrawals in Massive Open Online Courses. Research in these areas are usually carried out using deep learning to detect complex nonlinear patterns in students' learning sequences. However, the volatility and sparsity of data always weaken the performance of deep neural networks. Prevailing approaches always required an additional data smoothing or interpolation step independent of the prediction model, which may lose valuable information or introduce inauthentic data. Besides, when modeling the SDP problem as a binary classification task, previous works often required to specify an observation window, which may lead to inconsistent prediction results with different settings. To address these issues in an end-to-end learning framework, we propose a Survival Analysis based Volatility and Sparsity Modeling Network (SAVSNet). Particularly, SAVSNet smooths the volatile time series by convolution network while preserving the original data information using Long-Short Term Memory Network (LSTM). Furthermore, we propose a Time-Missing-Aware LSTM unit to mitigate the impact of data sparsity by integrating informative missingness patterns into the model. To achieve consistent predictions along time, a survival analysis loss fuction is adopted for parameter estimation and the model outputs monotonically decreasing survival probabilities. In the experiments, we compare the proposed method with state-of-the-art methods in two real-world MOOC datasets and the experiment results show the effectiveness of our proposed model.