Triple-E Principle: Leveraging Occam's Razor for Dance Energy Expenditure Estimation

Area of Science:

• Biomedical Engineering
• Sports Science
• Machine Learning

Background:

• Assessing energy expenditure in diverse physical activities like dance is challenging due to varied styles and tempos.
• Traditional methods using empirical formulas in accelerometers introduce significant biases.
• Existing wearable sensor solutions increase model complexity.

Purpose of the Study:

• To propose the Triple-E (Effectiveness, Efficiency, Extension) principle as a framework for developing state-of-the-art machine learning models for energy expenditure estimation in dance.
• To minimize model complexity and optimize sensor placement for accurate energy expenditure assessment.
• To validate the proposed framework using a cohort of participants engaged in various dance routines.

Main Methods:

• Recruited 250 participants (mean age: 63.0 ± 6.0 years) performing ballroom, aerobic, or square dance.
• Utilized ActiGraph wGT3X-BT accelerometers at five anatomical locations and CORTEX MetaMax 3B gas analyzer for metabolic data.
• Analyzed 311 physiological signal sequences and 1,555 acceleration count sequences.

Main Results:

• Empirical formulas showed high inaccuracy (MAPE > 50%, RMSE > 3.23) for dance energy expenditure.
• Bidirectional stepwise regression achieved a goodness-of-fit of 0.73, identifying optimal accelerometer sites.
• Random forest and deep learning models significantly reduced errors (RMSE as low as 0.15).
• Wrist accelerometers and heart rate alone offered sufficient accuracy, demonstrating a trade-off between Effectiveness and Efficiency.

Conclusions:

• Introduced a novel quantitative and unified model assessment system for energy expenditure estimation.
• Validated the effectiveness, efficiency, and extensibility of various machine learning models in the context of dance.
• Provided detailed insights into optimal model selection and sensor placement for accurate and efficient energy expenditure assessment.