Differentiating Human Falls from Daily Activities Using Machine Learning Methods Based on Accelerometer and Altimeter...

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Human activity recognition (HAR) is crucial for health monitoring and safety.
• Fall detection systems using wearable sensors are vital for elderly care and injury prevention.
• Integrating data from multiple sensors (sensor fusion) can improve HAR system accuracy.

Purpose of the Study:

• To analyze wearable sensor data for human activity recognition.
• To develop and evaluate machine learning models for distinguishing between activities of daily living (ADLs) and falls.
• To assess the effectiveness of sensor fusion in enhancing fall detection performance.

Main Methods:

• Utilized signal data from accelerometers and barometric altimeters.
• Implemented binary classification (ADLs vs. falls) and multiclass classification (running, walking, sitting, jumping, falling).
• Applied traditional machine learning models: Random Forest, Support Vector Machine, XGBoost, Logistic Regression, and Majority Voter.

Main Results:

• Combined features from accelerometers and barometric altimeters generally outperformed single-sensor models.
• Sensor fusion approaches demonstrated improved accuracy in both binary and multiclass activity recognition tasks.
• Machine learning models showed significant potential for reliable fall detection using integrated sensor data.

Conclusions:

• Sensor fusion significantly enhances the performance of human activity recognition systems for fall detection.
• Combining accelerometer and barometric altimeter data provides a robust foundation for advanced HAR applications.
• The study highlights the efficacy of traditional machine learning models in conjunction with sensor fusion for fall detection.