Single-accelerometer-based daily physical activity classification.

Area of Science:

• Biomedical Engineering
• Human Activity Recognition
• Wearable Technology

Background:

• Accurate human physical activity classification is crucial for assessing daily energy expenditure.
• Wearable sensors, like accelerometers, offer a non-intrusive method for activity monitoring.
• Developing robust classification models is essential for reliable energy expenditure estimations.

Purpose of the Study:

• To compare Bayesian classification with Decision Tree methods for human physical activity recognition.
• To evaluate the accuracy of classifying daily activities including walking, running, cycling, driving, and sports.
• To assess the feasibility of using a single waist-worn accelerometer for energy expenditure monitoring.

Main Methods:

• Utilized a single tri-axial accelerometer placed on the waist to collect acceleration data.
• Collected data from 24 subjects performing naturalistic daily activities.
• Applied Principal Component Analysis (PCA) for feature reduction and correlation removal.
• Compared Bayesian classification with a Decision Tree approach using leave-one-subject-out and 10-fold cross-validation.

Main Results:

• Achieved a classification accuracy of approximately 80% for the target physical activities.
• Demonstrated comparable performance between the Bayesian classifier and the Decision Tree classifier.
• Principal Component Analysis effectively reduced feature vector dimensions.

Conclusions:

• A single waist-worn accelerometer can achieve reliable human physical activity classification.
• The Bayesian classifier offers advantages in extensibility for future modifications and additions of activities.
• The developed methods show promise for practical applications in energy expenditure assessment.