This document describes research on using machine learning to predict student dropouts in online learning environments. It presents a new deep learning model called GRU-AE that is designed to handle the challenges of predicting dropouts in e-degree programs, which have longer student trajectories with potential gaps in activity. The GRU-AE model uses autoencoders to mitigate sparse and irregular student activity data, and gated recurrent units to learn relationships between activities over time. An experiment on real e-degree data found that GRU-AE improved prediction accuracy over other methods, especially for the longer sequences and sparser data that occur in e-degrees.