A Feature Engineering Method for Smartphone-Based Fall Detection.

Area of Science:

• Biomedical Engineering
• Computer Science
• Gerontology

Background:

• Falls are a leading cause of unintentional death globally, particularly among the elderly.
• Timely fall detection is critical for effective treatment and injury reduction.
• Existing methods require further development for accuracy and robustness.

Purpose of the Study:

• To develop and evaluate a smartphone-based fall detection system.
• To assess the performance of K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) classifiers.
• To ensure the method's effectiveness across diverse datasets and scenarios.

Main Methods:

• Utilized smartphone accelerometer data for fall event prediction.
• Employed K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) machine learning algorithms.
• Conducted rigorous evaluations on simulated (UniMiB SHAR, MobiAct) and real-world (FARSEEING) datasets, including cross-dataset validation.
• Performed SHAP-based interpretability analysis to understand feature influence.

Main Results:

• Achieved high accuracy in same-dataset evaluations (e.g., 98.45% on UniMiB SHAR, 99.89% on MobiAct).
• Demonstrated strong cross-dataset performance (up to 96.41% on MobiAct, 95.35% sensitivity and 98.12% specificity on FARSEEING).
• Identified key features influencing fall detection through interpretability analysis.

Conclusions:

• The proposed smartphone-based method is highly effective and robust for fall detection.
• The approach offers transparency in its decision-making process.
• This technology holds significant potential for improving elderly safety and reducing fall-related injuries.