Identifying at-risk workers using fNIRS-based mental load classification: A mixed reality study

Abstract

Construction is one of the most hazardous industries, in part because it involves dynamic and cognitively demanding tasks that tax workers' mental resources. Though some previous studies showed increases in workers' mental load can adversely affect the safety-critical function of workers' situational awareness, limited previous work has utilized machine learning (ML) techniques for mental load classification models with Functional Near-Infrared Spectroscopy (fNIRS) signals to passively identify at-risk workers suffering excessive mental loads. Within an immersive mixed-reality environment simulating an electrical construction task, this study employed ensemble modeling to classify workers' mental load via their fNIRS-captured brain activation. Beyond identifying the prefrontal and motor cortices as the most important brain areas contributing to ML-driven mental load classifications when using fNIRS, this paper also proposes a mental load classification methodology that synthesizes neuroimaging data with individuals' risk-propensity scores to address the limitations of existing personalized and generalized ML models. The results reveal better performance for fNIRS-based mental load classifications given participants' personal risk propensity (risk-seeker and risk-averse). This research, therefore, (1) offers valuable insights into the importance of identifying at-risk workers who are more likely to experience impaired cognitive processing and situational awareness due to increased mental load, and (2) sheds light on reliable pipelines for developing fNIRS-based mental load classification.