Multimodal Machine Learning for Student Retention Prediction in the College of Engineering

Integrating Temporal, Textual, and Tabular Features

Abstract

Student retention analysis and prediction supports interventions in higher education. We present a web-based tool to predict first-year and multi-year retention in the College of Engineering at Tennessee Technological University. The system integrates socio-demographic attributes, academic performance indicators, and advisement notes as predictive features. Structured and NLP-derived features are fused in a hybrid architecture with XGBoost one and multi-term predictions, respectively, with explainability using SHAP to identify influential factors for retention prediction.