Explainable Siamese Neural Networks for Detection of High Fall Risk Older Adults in the Community Based on Gait...

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• Falls in older adults are a major public health issue, causing injuries, increased healthcare costs, and reduced quality of life.
• Current fall risk assessment lacks definitive metrics, necessitating improved predictive capabilities for timely intervention.
• Accurate prediction of fall risk is crucial for implementing preventive strategies and maintaining independence in the elderly population.

Purpose of the Study:

• To develop a novel approach for fall risk assessment using biomechanical gait data.
• To transform time-series gait data into visual representations for deep learning (DL) application.
• To improve the accuracy and explainability of fall risk prediction models.

Main Methods:

• Utilized convolutional neural networks (CNNs) and Siamese neural networks (SNNs) for robust predictive modeling.
• Transformed biomechanical time-series gait data into visual representations.
• Employed Grad-CAM for generating class-discriminative activation maps to enhance model explainability.

Main Results:

• Achieved 83.29% accuracy in fall risk prediction using a random forest (RF) machine learning (ML) model.
• Successfully extracted distinctive gait-related features for improved prediction.
• Demonstrated enhanced explainability of the predictive model through visualization techniques.

Conclusions:

• Advanced computational tools and machine learning algorithms show significant potential for fall risk prediction.
• The proposed method can help reduce healthcare burdens associated with falls in older adults.
• Improved fall risk assessment can promote greater independence and well-being among the elderly population.