In Silico Diabetes Management and Prediction: A Personalized Hybrid Physics-ML System

Overview

Every 9 seconds, one person dies from diabetes, yet current technologies have significant limitations. I engineered a first-of-a-kind hybrid Physics-ML "glass-box" engine reconstructing each patient's unique metabolism. It uses Ordinary Differential Equations (ODEs) to model glucose flux (dG/dt), with gamma insulin pharmacokinetics curves, GI-weighted dual-wave absorption model, & circadian, exercise, stress & sleep modifiers. A Bayesian engine self-calibrates five parameters. Euler integration with 50 Monte Carlo simulations generates a 4-hour forecast with 80% confidence bands. Validated across four independent clinical datasets (N = 19,290; 2,858 patients), my system achieved mean bias = -0.15mg/dL, R² =0.896, AAE = 11.6mg/dL, MAPE = 7.0%, RMSE = 14.8mg/dL, paired t-test p = 0.15 > 𝛂, & Clarke A+B = 100% — confirming very strong retrospective clinical accuracy. My system also outperforms four recent peer-reviewed models — despite mine forecasting 4-8 times further ahead. It could prevent hypoglycemia/hyperglycemia before occurring — saving millions of lives.

Awards (2)

• Silver Medal
• Selected for CWSF 2026

Competition history

• CWSF 2026 Digital Technology

Related projects

Closest projects by meaning, across every fair and year in the corpus.