Comparative Analysis of Force-Sensitive Resistors and Triaxial Accelerometers for Sitting Posture Classification

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Sedentary behaviors and poor postures negatively impact health, especially for individuals with reduced mobility.
• Accurate posture detection is crucial for developing effective interventions and monitoring health status.
• Existing posture monitoring systems may lack accuracy, efficiency, or cost-effectiveness.

Purpose of the Study:

• To develop and evaluate a posture detection system using force-sensitive resistors (FSRs) and triaxial accelerometers.
• To compare the performance of various machine learning algorithms for posture classification.
• To identify the most effective sensor type and machine learning model for posture detection.

Main Methods:

• Utilized four force-sensitive resistors (FSRs) and two triaxial accelerometers, selected for consistency and linearity.
• Compared k-nearest neighbor (KNN), Decision Tree, Discriminant Analysis, Naive Bayes, and Support Vector Machine (SVM) algorithms.
• Optimized KNN hyperparameters, identifying the city block metric with K=3 as optimal.

Main Results:

• The k-nearest neighbor (KNN) algorithm outperformed other machine learning models in classification accuracy.
• Triaxial accelerometers achieved higher accuracy (99.4% training, 99.0% testing) compared to FSRs (96.6% training, 95.4% testing).
• Accelerometers also demonstrated slightly reduced processing times for both training and testing phases.

Conclusions:

• Triaxial accelerometers are more effective than FSRs for posture detection due to higher accuracy and efficiency.
• The developed KNN-based system offers a cost-effective and compact solution for posture monitoring.
• This technology has significant potential for aiding individuals with mobility impairments and promoting healthier postures.