An Individualized Machine Learning Approach for Human Body Weight Estimation Using Smart Shoe Insoles.

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Machine Learning

Background:

• Rapid weight fluctuations signal serious health issues like heart failure or dehydration.
• Daily weigh-ins fail to capture intra-day weight changes and suffer from poor patient compliance.
• Current monitoring methods lack accuracy and real-time capabilities for continuous weight assessment.

Purpose of the Study:

• To develop a machine learning framework for continuous, real-time body weight estimation using in-shoe sensors.
• To address the scarcity of public datasets for training weight estimation models by introducing two novel datasets.
• To create an adaptive system accounting for individual patient factors like footwear and gait.

Main Methods:

• Utilized machine learning models to predict continuous body weight from data collected by shoe insole sensors.
• Developed and validated a novel framework incorporating patient-specific parameters (shoe type, posture, foot shape, gait).
• Introduced two new datasets to facilitate the training and benchmarking of weight estimation models.

Main Results:

• Achieved high accuracy in weight estimation, with Mean Absolute Percentage Errors of 0.61% (less controlled) and 0.74% (more controlled).
• Demonstrated low Mean Absolute Errors of 1.009 lbs. (less controlled) and 1.154 lbs. (more controlled).
• The proposed framework shows significant potential for reliable patient monitoring.

Conclusions:

• Shoe insole sensors coupled with machine learning offer a promising solution for accurate, real-time body weight monitoring.
• This technology can enhance telehealth services and support reliable aging-in-place monitoring.
• The developed framework and datasets pave the way for advanced applications in remote patient care.