Application of Machine Learning Approach to Classify Human Activity Level Based on Lifelog Data.

Area of Science:

• Biomedical Engineering
• Health Informatics
• Machine Learning

Background:

• Wearable devices generate extensive patient lifelog data, including physiological metrics.
• Accurate classification of human activity levels is crucial for personalized healthcare and monitoring.

Purpose of the Study:

• To develop and evaluate machine learning models for classifying human activity levels.
• To utilize patient lifelog data from wearable devices for activity recognition.

Main Methods:

• Collected heart rate, step count, and calorie consumption data from 182 patients over two months.
• Pre-processed integrated time-series wearable data (80% training, 20% testing).
• Evaluated 16 algorithms (12 traditional ML, 4 deep learning) using 5-fold cross-validation and parameter optimization.

Main Results:

• Machine learning models achieved high accuracy in classifying human activity levels.
• Classification performance was particularly strong using heart rate and step count data.
• Optimized models demonstrated effectiveness on new patient data.

Conclusions:

• Human activity level classification is feasible and accurate using wearable sensor data.
• Machine learning, especially with heart rate and step count, offers a robust approach for patient activity monitoring.
• This model has potential applications in remote patient care and health management.