A universal hybrid decision tree classifier design for human activity classification.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human Activity Recognition

Background:

• Reliable daily life activity classification is crucial for effective patient treatment and rehabilitation.
• Existing methods may lack adaptability and accuracy for personalized patient monitoring.

Purpose of the Study:

• To develop and evaluate a universal hybrid decision tree classifier for accurate daily life activity recognition.
• To compare the proposed classifier's performance against personalized machine learning models.

Main Methods:

• Implementation of a hybrid decision tree classifier adaptable from population models with individual training data.
• Utilizing 14 triaxial accelerometers to monitor seven subjects performing 14 distinct daily activities.
• Employing leave-one-out cross-validation for performance assessment.

Main Results:

• The proposed hybrid decision tree classifier achieved an average accuracy of 89.9%.
• This significantly outperformed MATLAB's personalized tree classifiers, which yielded 69.2% accuracy.
• The classifier demonstrated flexibility in implementing diverse decision rules.

Conclusions:

• The universal hybrid decision tree classifier offers a highly accurate and adaptable solution for human activity recognition.
• This technology has strong potential for enhancing remote patient monitoring, chronic disease management, and rehabilitative therapies.