Impact of Gait Parameters and Their Variability on Fall Risk Assessment Accuracy Using Wearable Sensor

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Wearable sensors provide precise gait data for fall risk assessment.
• The influence of gait variability on fall risk prediction models is not fully understood.

Purpose of the Study:

• To evaluate machine learning models for predicting fall risk in frail older adults using gait parameters.
• To determine optimal walking test durations for fall risk assessment with wearable sensors.

Main Methods:

• Trained logistic regression, support vector machines (SVM), and artificial neural network models on gait data from 163 frail older adults.
• Collected gait parameters and variability using a foot-mounted inertial measurement unit (IMU).
• Utilized leave-one-out cross-validation and walking test durations from 1 to 15 minutes.

Main Results:

• Optimal walking test durations for fall risk prediction ranged from 6 to 10 minutes.
• The artificial neural network model achieved the highest prediction accuracy (0.96) with an 8-minute test duration.
• Gait parameters and their variability derived from wearable sensors are effective predictors of fall risk.

Conclusions:

• Machine learning models, particularly artificial neural networks, can accurately predict fall risk in older adults using wearable sensor data.
• Walking test durations between 6 and 10 minutes are optimal for fall risk assessment using IMU-based gait analysis.
• Findings inform the design of clinical fall risk assessment protocols utilizing wearable technology.