EFFECT OF NOURISHING NUTRIENTS ON POSTPRANDIAL GLUCOSE RESPONSE ON TYPE 1 DIABETES THROUGH FEED-FORWARD NEURAL NETWORK

Abstract

Type 1 Diabetes (T1D) management relies heavily on predicting Postprandial Glucose Response
(PGR) for insulin dosing, crucial for patient well-being. While carbohydrates traditionally dominate PGR
prediction, other nutritional factors significantly influence Blood Glucose Levels (BGLs). Leveraging Machine
Learning (ML), this study investigates the impact of carbohydrates, proteins, lipids, fibers, and energy intake on
short to middle-term BGL prediction in T1D patients using Artificial Pancreas (AP) systems. A Feed-Forward
Neural Network (FFNN) incorporating insulin doses, blood glucose, and nutritional factors predicts BGLs at 15,
30, 45, and 60 minutes post-meal. Both public and self-produced data validate the model. Further extending
beyond traditional ANN models, ensemble techniques including FFNN, MLP, Bagging Classifier with Random
Forest (RF), and Voting Classifier (combining Bagging Classifier with RF and Decision Tree) were explored.
Ensemble methods significantly enhance performance, potentially achieving above 95% accuracy. This research
underscores the importance of considering diverse nutritional factors for accurate postprandial BGL predictions,
advancing personalized T1D management with AP systems