Fusion of Clinical, Self-Reported, and Multisensor Data for Predicting Falls

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls are a major cause of reduced mobility and independence in the elderly.
• Global population aging necessitates advanced fall prediction strategies.
• Current methods often lack multifactorial data integration.

Purpose of the Study:

• To develop and evaluate multifactorial data fusion approaches for predicting falls in community-dwelling older adults.
• To compare the efficacy of early, late, and slow fusion techniques.
• To identify the optimal fusion strategy for maximizing fall prediction recall.

Main Methods:

• A multifactorial screening protocol was applied to 281 adults over 65.
• Clinical, self-reported, and instrumented functional test data (inertial sensors, pressure platform) were collected.
• Early, late, and slow data fusion approaches were implemented and compared using a classification pipeline and multilayer perceptron.

Main Results:

• All three fusion approaches (early, late, slow) demonstrated comparable performance across most metrics.
• The late fusion approach achieved a superior recall rate ([Formula: see text]) when prioritizing recall over specificity.
• Feature selection and classifier optimization were performed using grid search with cross-validation.

Conclusions:

• Multifactorial data fusion is effective for predicting falls in older adults.
• Late fusion offers an advantage when high sensitivity (recall) is the primary goal in fall prediction.
• This research contributes to developing proactive strategies for fall prevention in aging populations.