Investigating Performance of an Embedded Machine Learning Solution for Classifying Postural Behaviors.

Area of Science:

• Biomedical Engineering
• Neurology
• Machine Learning

Background:

• Postural instability is a critical concern in degenerative diseases, often predicting falls.
• Monitoring postural behavior is essential for patient care and fall prevention.

Purpose of the Study:

• To present a multi-layer perceptron (MLP) approach for classifying four distinct postural behaviors.
• To evaluate the robustness of the MLP methodology against noisy data and varying feature sets.
• To compare the MLP approach with threshold-based algorithms.

Main Methods:

• Implementation of an embedded sensing architecture.
• Classification of postural behaviors using a multi-layer perceptron (MLP).
• Investigation of data robustness with varying noise levels and feature sets.

Main Results:

• The MLP approach achieved a reliability index near 100% even with noisy input data.
• A negligible performance drop (less than 5%) was observed across investigated noise levels.
• The methodology demonstrated superior robustness compared to threshold-based algorithms.

Conclusions:

• The developed MLP approach offers a highly reliable method for classifying postural behaviors.
• This technique is robust against noisy data, making it suitable for real-world applications.
• The findings suggest a significant advancement over existing threshold-based methods for postural instability monitoring.