MACHINE LEARNING OF ELECTROENCEPHALOGRAPHY SIGNALS AND EYE MOVEMENTS TO CLASSIFY WORK-IN-PROGRESS RISK AT CONSTRUCTION SITES

Abstract

The construction industry has consistently faced high accident rates and delays in recognizing hazards, posing significant risks to onsite personnel. Traditional hazard detection methods are often reactive rather than proactive, emphasizing a pressing need for innovative solutions. Despite advances in safety technology, a considerable gap remains in real-time, accurate hazard recognition at construction sites. Current technologies do not fully leverage physiological data to predict and mitigate risks. This research introduces a groundbreaking approach by employing machine learning to analyze electroencephalography (EEG) signals and eye movement data, enabling real-time differentiation of safe, warning, and hazardous visual cues. A Random Forest model with an impressive classification accuracy of 99.04% has been developed, marking a significant enhancement in identifying potential hazards. The possible impact of integrating EEG and eye movement analyses into wearable devices or onsite sensors is substantial, as it could revolutionize safety protocols in the construction industry, fostering a safer future.