Machine Learning-Based Framework for Pre-Impact Same-Level Fall and Fall-from-Height Detection in Construction Sites...

Area of Science:

• Occupational Safety and Health
• Biomedical Engineering
• Machine Learning Applications

Background:

• Same-level-falls (SLFs) and falls-from-height (FFHs) are significant causes of severe injuries and fatalities in construction.
• Accurate detection of these falls is crucial for developing effective fall-prevention systems, but current methods struggle with false positives in dynamic construction environments.
• Wearable fall-prevention devices require rapid and precise pre-impact detection capabilities to be effective.

Purpose of the Study:

• To establish a machine learning-based approach for accurate identification of SLFs, FFHs, and non-fall events.
• To utilize data from a single waist-mounted inertial measurement unit (IMU) for fall detection.
• To develop a system capable of rapid, pre-impact detection suitable for wearable fall-prevention technologies.

Main Methods:

• Collected data from 48 participants performing various non-fall activities, SLFs, and FFHs, using a dummy for higher falls.
• Employed a two-stage feature extraction process to generate 168 descriptors per data window.
• Utilized an ensemble SHAP-PFI method to select the 153 most informative variables and a weighted XGBoost classifier optimized via Bayesian techniques.

Main Results:

• The optimized XGBoost classifier achieved a high average macro F1-score of 0.901 and macro Matthews correlation coefficient of 0.869.
• The system demonstrated low latency (1.51 × 10^-3 ms per window) and achieved average lead times of 402 ms for SLFs and 640 ms for FFHs.
• These lead times significantly exceed the 130 ms inflation time required for wearable airbags, indicating effective pre-impact detection.

Conclusions:

• The developed machine learning approach provides rapid and precise detection of SLFs and FFHs using a single IMU.
• This pre-impact detection capability positions the system as a viable core component for advanced wearable fall-prevention devices.
• The findings offer a promising solution for enhancing worker safety in high-risk construction environments.