Machine Learning-Based Blood Glucose Spike Prediction: Intelligent Health Decision-Making Using Nutrition and Lifestyle Data

A machine learning project developed by NutriGlyc AI Solutions that predicts blood glucose spikes using nutrition, health, and lifestyle data to support diabetes prevention and personalized nutrition management.

Project Source: GitHub Repository (Author: susandangana, Original Title: glucose_spike_prediction, Link: https://github.com/susandangana/glucose_spike_prediction, Release Date: 2026-06-13).

Core Technologies: Implemented using algorithms like Logistic Regression, based on Python and scikit-learn, aiming to provide data support for intelligent health decision-making.

Type 2 diabetes has become a major global public health challenge, with rising incidence rates caused by poor diet, sedentary lifestyles, etc. Traditional health assessments rely on manual work, which is time-consuming and lacks predictive ability, making it difficult to identify high-risk groups in a timely manner.

NutriGlyc AI Solutions focuses on health technology and nutritional analysis. This project uses AI models to predict blood glucose spikes, aiding diabetes prevention and personalized nutrition management, which is a concrete practice of its mission.

The project dataset covers multi-dimensional variables:

• Demographic features: Age, gender, BMI, type of diabetes
• Nutritional intake indicators: Carbohydrate, protein, fat, fiber, sugar intake, Glycemic Index (GI), Glycemic Load (GL)
• Lifestyle factors: Physical activity level, stress level, smoking status, medication adherence
• Medical indicators: Insulin dosage

The target variable is a binary "blood glucose spike" (0 = none, 1 = present), and the feature design reflects the multi-factor interactivity of blood glucose regulation.

The project follows a structured workflow: Data Cleaning → EDA → Feature Engineering → Multicollinearity Assessment → Permutation Importance Feature Selection → Model Development and Comparison → Hyperparameter Tuning → Final Evaluation.

Five algorithms were compared: Logistic Regression, Random Forest, XGBoost, SVM, and KNN. Logistic Regression was finally selected because it achieves the best balance between interpretability and predictive performance.

Model Performance Metrics:

Metric	Score
Accuracy	76.0%
Precision	72.1%
Recall	78.6%
F1 Score	75.2%
ROC-AUC	0.848

Key Influencing Factors (by importance): Carbohydrate intake, carbohydrate-to-fiber ratio, glycemic load, fiber intake, physical activity, stress level.

Dietary Recommendations: Monitor carbohydrate intake; increase fiber in meals; prioritize low glycemic load foods.

Lifestyle Recommendations: Regular physical activity; effective stress management.

Tech Stack: Python ecosystem tools (Pandas, NumPy for data processing; Matplotlib, Seaborn for visualization; Scikit-Learn, XGBoost for machine learning; Joblib for model persistence).

Project Value: Provides a complete ML project template for developers; offers AI-assisted decision-making tools for medical practitioners; delivers evidence-based health recommendations for users.

Future Outlook: Incorporate continuous glucose monitoring and wearable device data; integrate into mobile apps/health platforms to achieve more refined and personalized predictions.