Sensor-based evaluation of intermittent fasting regimes: a machine learning and statistical approach.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Nutritional Science

Background:

• Dietary adherence is crucial for health outcomes, especially in intermittent fasting (IF).
• Objective monitoring of IF adherence is challenging using traditional methods.
• Sensor technology offers a promising avenue for passive dietary assessment.

Purpose of the Study:

• To develop and evaluate models for assessing dietary adherence in intermittent fasting using sensor data.
• To compare the performance of machine learning and statistical methods for classifying fasting states.
• To create a user-friendly dashboard for visualizing adherence data.

Main Methods:

• Utilized time-series data from human trials including continuous glucose monitoring, acceleration, and food diaries.
• Incorporated a synthetic dataset for model training and validation.
• Applied machine learning models and the Hutchison Heuristic statistical method.
• Developed a results visualization dashboard.

Main Results:

• Machine learning models achieved a high average F1-score of 0.88 in distinguishing fasting vs. non-fasting periods.
• The Hutchison Heuristic method demonstrated robustness across diverse cohorts, including those with type 1 diabetes.
• The developed dashboard provided efficient and user-friendly data visualization.

Conclusions:

• Sensor data, coupled with advanced statistical and machine learning techniques, effectively enables passive evaluation of dietary adherence in intermittent fasting.
• These methods show high reliability in classifying fasting states.
• The approach is applicable across different populations, including those with diabetes.