Machine Learning-Based Pre-Impact Fall Detection Model to Discriminate Various Types of Fall

Area of Science:

• Biomedical Engineering
• Machine Learning in Healthcare
• Gerontology

Background:

• Pre-impact fall detection is crucial for timely alerts, reducing hospitalization risks and severe injuries.
• Current models often lack the ability to classify diverse fall types or activities before impact.

Purpose of the Study:

• To develop and evaluate a robust multiclass pre-impact fall detection model.
• To compare the novel model's performance against existing fall detection systems.

Main Methods:

• Utilized an inertial measuring unit to collect accelerometer data from twelve healthy subjects performing various activities.
• Developed a novel model incorporating feature calculation, infinite latent feature selection (ILFS), auto-labeling, and machine learning classifiers for time series data.
• Applied nine machine learning classifiers for pre-impact fall detection and results sorting.

Main Results:

• The proposed model achieved superior multiclass classification accuracy: 99.57% ± 0.01% for discrete data and 99.84% ± 0.02% for continuous time series data.
• Demonstrated significantly higher accuracy compared to previous models (p < 0.01).

Conclusions:

• The developed multiclass pre-impact fall detection model offers high accuracy in classifying falls and activities.
• Future applications include integration into fall protector devices for enhanced fall prevention and immediate feedback mechanisms.