Machine learning algorithms based on signals from a single wearable inertial sensor can detect surface- and...

Area of Science:

• Biomechanics
• Machine Learning
• Wearable Technology

Background:

• Walking patterns change with age and surface conditions.
• Detecting these changes can help identify individuals at risk of falls.
• Inertial Motion Units (IMUs) offer a portable method for capturing gait data.

Purpose of the Study:

• To determine if a machine learning algorithm can differentiate between age groups and walking surfaces using IMU data.
• To evaluate the effectiveness of different sensor data inputs for this detection task.

Main Methods:

• Seventeen older and eighteen young healthy adults walked on flat and uneven surfaces.
• An IMU sensor placed at the L5 vertebra collected triaxial accelerometer, gyroscope, and magnetometer data.
• A deep learning network with long short-term memory units was trained and tested on segmented IMU data.

Main Results:

• The fully trained machine learning model, using all sensor data, achieved high accuracy (96.3% for surface, 94.7% for age).
• This comprehensive model outperformed models using individual sensor data alone.
• The model demonstrated strong performance metrics including AUC (0.97 for surface, 0.96 for age), precision, recall, and f1-score.

Conclusions:

• Machine learning algorithms processing IMU signals can effectively detect age-related and surface-related variations in walking.
• This approach holds potential for early identification of fall risk and intervention strategies.