Assessing neural markers of attention during exposure to construction noise using machine learning classification of electroencephalogram data

Abstract

Construction noise is one of the leading causes of attention impairment both for workers within construction sites and individuals in their direct vicinity. Distraction caused by construction noise can significantly affect productivity of people in the surrounding area. In addition, lack of adequate attention by workers in construction sites can increase the risk of mistakes potentially increasing work-related accidents. This study highlights the feasibility of using electroencephalogram (EEG) data for detecting distractions caused by construction noise. EEG data were collected from 23 participants while they completed a standard attention test (Go-NoGo) during exposure to construction noise. Both frequency-based features and non-linear features were extracted from the EEG signal and subject-independent machine learning models were applied. The results showed that channel FC5, with an F1 score of 68.41 % and an accuracy score of 70.28 % presented the best performance among all channels and the defined channel clusters. These findings will help inform the best features and channels to select for developing new purpose-built EEG devices with a smaller number of channels that are suited for distraction detection due to noise exposure in construction sites.