Fall prediction in a quiet standing balance test via machine learning: Is it possible?

Area of Science:

• Biomechanics and Machine Learning
• Gerontology and Fall Prevention

Background:

• The global elderly population is rapidly expanding, leading to an increased incidence of falls.
• Current clinical gait and posture assessments lack the sensitivity and specificity required for accurate fall risk prediction.
• Machine learning offers a novel approach to analyze complex data for improved outcome predictions.

Purpose of the Study:

• To evaluate the performance of machine learning algorithms in classifying participants by age and fall history using stabilometric data.
• To determine if machine learning can identify individuals at high risk of falling based on gait and posture features.

Main Methods:

• Utilized a public database of stabilometric assessments from 163 participants (aged 18-85).
• Applied six machine learning algorithms: Logistic Regression, Linear Discriminant Analysis, K Nearest-neighbours, Decision Tree Classifier, Gaussian Naive Bayes, and C-Support Vector Classification.
• Extracted features from demographic, sociocultural, and health status information.

Main Results:

• All tested machine learning models successfully classified participants into 'young' or 'old' age groups.
• No machine learning model demonstrated the ability to accurately identify participants at high risk of falling.
• Static posturography, likely due to low specificity in daily living activities, was insufficient for predicting fall risk.

Conclusions:

• Machine learning algorithms can effectively differentiate age groups based on stabilometric data.
• Static posturography is not a reliable method for predicting fall risk in the elderly.
• Future research should investigate dynamic posturography for more accurate fall risk assessment.