This paper investigates the efficacy of various deep learning architectures in predicting the onset and progression of diabetic complications using the "Diabetic 11" longitudinal dataset. By integrating demographic, clinical, and biochemical markers over 11 distinct time intervals or patient clusters, we propose a novel transformer-based model that outperforms traditional RNNs in early risk detection.
Compare Random Forests, Gradient Boosting (XGBoost), and LSTM networks for classification accuracy. 3. Methodology
Below is a proposal for a high-impact paper using this data: