Location
https://www.kennesaw.edu/ccse/events/computing-showcase/sp25-cday-program.php
Streaming Media
Event Website
http://ebelfarsi.com/glucose_response_predictor
Document Type
Event
Start Date
15-4-2025 4:00 PM
Description
This study predicts postprandial glucose peaks and spike durations using 10-day multimodal data from 10 participants. Glucose, meals, workouts, and insulin doses were logged via the Nutrilyzer web app. Macronutrient content carbs, fats, and proteins was extracted using GPT-Vision, a highly accurate food analysis tool. These tuples were normalized to baseline glucose and aligned with a 3-hour window post-meal. Three models were tested: LSTM, Time Series Transformer, and ARIMA. LSTM performed best with 83.78% accuracy, followed by Transformer (71.43%) and ARIMA (62.41%). Results show the promise of AI-based food logging and time series modeling for personalized glucose forecasting.
Included in
GRP-088 Nutrilyzer: A Vision-Based App for Macronutrient Estimation and Blood Glucose Response Prediction
https://www.kennesaw.edu/ccse/events/computing-showcase/sp25-cday-program.php
This study predicts postprandial glucose peaks and spike durations using 10-day multimodal data from 10 participants. Glucose, meals, workouts, and insulin doses were logged via the Nutrilyzer web app. Macronutrient content carbs, fats, and proteins was extracted using GPT-Vision, a highly accurate food analysis tool. These tuples were normalized to baseline glucose and aligned with a 3-hour window post-meal. Three models were tested: LSTM, Time Series Transformer, and ARIMA. LSTM performed best with 83.78% accuracy, followed by Transformer (71.43%) and ARIMA (62.41%). Results show the promise of AI-based food logging and time series modeling for personalized glucose forecasting.
https://digitalcommons.kennesaw.edu/cday/Spring_2025/PhD_Research/7
