Optimizing Fall Risk Diagnosis in Older Adults Using a Bayesian Classifier and Simulated Annealing

Area of Science:

• Gerontology
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Fall risk assessment is crucial for preventing injuries in older adults.
• Current diagnostic tools can be difficult to apply in clinical settings.
• Improving the accuracy and explainability of fall risk classifiers is an ongoing challenge.

Purpose of the Study:

• To enhance the diagnostic capabilities of fall risk classifiers.
• To apply a Bayesian approach combined with the Simulated Annealing (SA) algorithm for feature selection.
• To optimize feature sets for improved classifier performance and interpretability.

Main Methods:

• Analyzed 47 features from 181 records, including Center of Pressure (CoP) indices and patient variables.
• Employed a wrapper method for feature selection using the SA algorithm.
• Optimized a cost function based on Area Under the Curve (AUC) differences and used a stratified 60-20-20% data split.

Main Results:

• The highest performance (0.815 ± 0.110) was achieved with 31 features.
• A reduced set of 15 features yielded comparable performance (0.780 ± 0.055) with significantly lower variability and higher explainability.
• The SA algorithm proved effective in selecting optimal feature subsets.

Conclusions:

• The Simulated Annealing algorithm is a valuable tool for feature selection in fall risk diagnosis.
• This approach offers an alternative or complementary method to existing clinical tools.
• Optimized feature selection can lead to more reliable and interpretable fall risk assessment.