College Student Retention Risk Analysis from Educational Database Using Multi-Task Multi-Modal Neural Fusion

We develop a Multimodal Spatiotemporal Neural Fusion network for MTL (MSNF-MTCL) to predict 5 important students' retention risks: future dropout, next semester type of duration dropout and cause dropout. First, we general purpose multi-modal neural fusion model MSNF learning academic information representation by fusing spatial temporal unstructured advising notes with spatiotemporal structured data. combines Bidirectional Encoder Representations from Transformers (BERT)-based document embedding framework represent each note, Long-Short Term Memory (LSTM) note embeddings, LSTM performance variables static demographics altogether. The final fused has been utilized on Multi-Task Cascade Learning (MTCL) towards building MSNF-MTCL predicting student risks. evaluate large educational database consists 36,445 college students over 18 years period time that provides promising performances comparing the nearest state-of-art models. Additionally, test fairness such given existence biases.