Machine Learning Algorithms for Activity-Intensity Recognition Using Accelerometer Data

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Health Informatics

Background:

• Activity recognition is crucial for pervasive healthcare monitoring.
• Intensity recognition, a key contextual parameter, is less explored.
• Accelerometer data can provide insights into daily activities.

Purpose of the Study:

• To investigate the advantage of coupling activity and intensity recognition (Activity-Intensity) using accelerometer data.
• To compare two supervised classification approaches: single classifier vs. separate classifiers.
• To evaluate the performance of k-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF) algorithms.

Main Methods:

• Collected accelerometer data to capture daily activities.
• Implemented a single classifier approach for joint Activity-Intensity recognition.
• Implemented a two-classifier approach for separate activity and intensity recognition.
• Utilized KNN, SVM, and RF algorithms for classification tasks.

Main Results:

• Both single and separate classifier approaches demonstrated viability for Activity-Intensity recognition.
• The single KNN classifier achieved 79% accuracy for coupled Activity-Intensity recognition.
• The separate classifier approach yielded 97% accuracy for activity (RF) and 80% for intensity (KNN), resulting in 78% for coupled recognition.

Conclusions:

• Coupling activity and intensity recognition enhances the description of daily activities.
• The single KNN classifier approach is effective for Activity-Intensity recognition.
• Findings support the development of decision systems for health professionals to improve movement evaluation.