AI-Driven Fall Prediction across Generations: Integrating Deep Learning and Machine Learning for Young, Middle-Aged,...

Area of Science:

• Gerontology and Public Health
• Artificial Intelligence in Healthcare
• Biomedical Informatics

Background:

• Falls pose a significant public health risk across all age groups.
• Existing artificial intelligence (AI) models for fall risk prediction lack comparative performance data and clarity on applicability to younger populations.
• This study addresses the need for robust, age-inclusive fall risk prediction models.

Purpose of the Study:

• To develop and compare various machine learning (ML) and deep learning (DL) models for fall risk prediction.
• To identify key predictive features for falls across different adult age groups.
• To evaluate model performance and generalizability in a diverse population.

Main Methods:

• Trained and evaluated five ML models (KNN, RF, GBDT, XGBoost, CatBoost) and two DL models (GRU, AGRU) on data from 1,441 community-dwelling adults.
• Utilized accuracy, precision, recall, F1 score, and AUROC for model evaluation.
• Conducted age-stratified analyses (20-45, 46-65, >65 years) and interpreted feature importance using SHapley Additive exPlanations.

Main Results:

• The AGRU deep learning model demonstrated superior performance with 91.39% accuracy and 0.934 AUROC overall.
• Pulse rate, living alone, systolic blood pressure, 5-times Sit-to-Stand test, and sex were identified as major fall predictors in the general population.
• Top predictive factors for fall risk differed significantly across the analyzed age strata.

Conclusions:

• A robust and interpretable deep learning model (AGRU) for fall risk identification across age groups was developed.
• Age-specific risk factors underscore the importance of personalized fall prevention strategies.
• External validation showed moderate generalizability, highlighting the need for larger, diverse datasets and sensor-based data integration for clinical application.