Domain-informed weight forecasting: leveraging behavioral and physiological sequences from wearables.

Area of Science:

• Computational health informatics
• Machine learning for health monitoring
• Sequence modeling for physiological data

Background:

• Existing body-weight forecasting models often require contemporaneous body weight or BMI, limiting practical use.
• Frequent weigh-ins are not always available, necessitating alternative forecasting approaches.
• Short-term body-weight forecasting can support personalized weight management.

Purpose of the Study:

• To develop a direct multi-step forecasting framework for predicting 7-day body-weight trajectories.
• To utilize behavioral, physiological, and lifestyle variables as inputs, excluding current body weight and BMI.
• To evaluate the framework's performance on both synthetic and real-world datasets.

Main Methods:

• Developed a Long Short-Term Memory (LSTM) based forecasting framework.
• Utilized the synthetic FitLife360 dataset for model development and benchmarking.
• Validated the framework on real-world wearable/lifelogging data (PMData) from 16 participants.
• Compared LSTM performance against Random Forest and XGBoost models.

Main Results:

• The proposed LSTM framework achieved the best overall performance on the synthetic dataset.
• Consistent performance gains were observed in feature-ablation analyses.
• The framework demonstrated feasibility and operational capability on real-world wearable/lifelogging records.

Conclusions:

• Domain-informed sequence modeling is feasible for short-horizon body-weight forecasting.
• The study provides a methodological foundation for future validation in larger cohorts.
• Short-horizon predictions reflect transient variability and should not be directly interpreted as adiposity change or cardiometabolic risk.