A Resource-Efficient Load Cell-Based Smart Bed System for Posture Classification.

Area of Science:

• Biomedical Engineering
• Healthcare Technology
• Machine Learning in Medicine

Background:

• Smart bed systems are crucial for patient monitoring and pressure ulcer prevention in healthcare.
• Existing vision- and biosignal-based methods raise privacy concerns and cause sensor discomfort.

Purpose of the Study:

• To develop a contact-free, minimally instrumented smart bed system for posture classification.
• To offer a cost-effective and efficient alternative to high-resolution pressure sensor arrays.

Main Methods:

• Load cells installed on bed legs measured pressure distribution for posture classification.
• Machine learning models (Random Forest, SVM, kNN) were trained on six predefined postures.
• User-specific data (height, weight) and five-fold cross-validation enhanced model generalization.

Main Results:

• The Random Forest model achieved the highest accuracy (80.49%), sensitivity (80.49%), specificity (82.09%), and F1-score (81.24%).
• Semi-lateral postures showed higher misclassification rates due to similar weight distribution.
• The proposed system demonstrated accurate, efficient, and privacy-preserving posture classification.

Conclusions:

• The non-contact smart bed framework is a viable alternative to high-resolution sensor arrays for posture classification.
• Future work includes expanding dataset diversity, dynamic posture transitions, and IoT integration.