Modeling physical activity data using L₀ -penalized expectile regression.

Area of Science:

• Biomedical Engineering
• Data Science
• Physical Activity Research

Background:

• Accelerometers are crucial for objective physical activity assessment.
• Current methods like cut points often ignore underlying activity patterns.
• Hidden Markov models (HMMs) offer an improvement but can be further enhanced.

Purpose of the Study:

• To introduce and evaluate a novel expectile regression approach for accelerometer data analysis.
• To improve the capture of physical activity intensity levels and patterns.
• To compare the new method against existing HMMs and cut point techniques.

Main Methods:

• Utilized expectile regression with a Whittaker smoother and an L0-penalty.
• Simulated 1,000 days of accelerometer data with 1 and 5-second epochs.
• Compared performance based on misclassification rate, bout identification, and level accuracy.

Main Results:

• Expectile regression demonstrated superior performance over HMMs and cut points.
• The method effectively distinguished between monotonous and variable activity patterns.
• Accurate estimation of activity intensity levels was achieved.

Conclusions:

• Expectile regression with a Whittaker smoother and L0-penalty is a promising advancement for accelerometer data modeling.
• This approach offers more nuanced and accurate physical activity intensity classification.
• The method holds potential for improving objective physical activity monitoring.