Data correction for seven activity trackers based on regression models.

Area of Science:

• Wearable technology
• Biomedical engineering
• Human activity recognition

Background:

• Activity trackers measure steps and energy expenditure (EE) for fitness.
• Accurate EE estimation is challenging due to sensor limitations and required personal data.
• Existing trackers often lack precision in EE and step measurements.

Purpose of the Study:

• To evaluate regression-based models for improving step count and EE estimation accuracy.
• To assess the effectiveness of these models across different activity types and devices.
• To determine the feasibility of low-cost implementation for enhanced tracker accuracy.

Main Methods:

• Collected data from seven activity trackers and two reference devices on 20 volunteers.
• Tested devices during office work, general activity, and walking.
• Developed and applied regression models using reference data to adjust tracker outputs.
• Statistically compared the relative percentage errors of original and adjusted values.

Main Results:

• Energy expenditure (EE) measurement accuracy improved across all tested devices during walking.
• Step count accuracy improved in five out of seven devices during walking.
• Regression models demonstrated effectiveness in enhancing measurement precision.
• Improvements were achieved with a low-cost, data-fitting approach.

Conclusions:

• Regression models offer a viable, low-cost method to enhance the accuracy of activity trackers.
• EE and step count precision can be significantly improved, particularly during walking activities.
• Implementation of these models within apps or back-end services is practical and effective.