The Key Factors in Physical Activity Type Detection Using Real-Life Data: A Systematic Review

Area of Science:

• Biomedical Engineering
• Kinesiology
• Wearable Technology

Background:

• Physical activity (PA) is crucial for health, yet understanding its types and impact on well-being is limited.
• Current PA classification methods often lack validity in real-life conditions.
• Real-life study designs are essential for accurate PA monitoring.

Purpose of the Study:

• To systematically review methods for physical activity type detection (PATD) in real-life settings.
• To focus on accelerometer data from portable devices for PATD.
• To identify challenges and propose recommendations for PATD in real-life environments.

Main Methods:

• Systematic literature review of 1,170 identified publications.
• Inclusion criteria: PATD methods, accelerometer data, real-life settings.
• Analysis of data collection, preprocessing, and PATD methods.

Main Results:

• 21 publications were included, organized by PATD process elements.
• High classification accuracies reported, but varied protocols hinder comparison.
• Minimal sensor configuration: two 3D accelerometers (thigh, hip) at 20 Hz for posture/motion activities.
• Decision trees are common classifiers for real-life data.

Conclusions:

• Fewer studies focus on PA types compared to intensity, especially in real-life settings.
• Standardization of data collection and performance reporting is urgently needed.
• Development of labeled, openly available reference datasets with a common evaluation framework is critical.