Maritime Shift Workers Sleepiness Detection System With Multi-Modality Cues

• Published in: IEEE access 2019, Vol.7, p.98792-98802
• Main Authors: Balandong, Rodney P., Tang, Tong Boon, Short, Michelle A., Saad, Naufal M.
• Format: Article
• Language: English
• Subjects: Bayes methods; Bayesian analysis; Bayesian network; Biomedical monitoring; classification; Conditional probability; Industries; Likelihood ratio; Monitoring; Schedules; Sleep; sleep quality; sleepiness detection; Support vector machines; Transport safety
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2929066

Sleepiness has been recognized as a causal factor in many round-the-clock industries. While individuals can subjectively express their momentary sleepiness level, sleepiness-related contextual factors (CF) can influence their perception of sleepiness and cognitive performance. In this paper, the self-reported sleepiness value (vSRS) was improved by transforming it into a kernel density estimate and the assignment of the class's score is done using a likelihood ratio test (IvSRS). We integrated multiple CF and IvSRS to model sleepiness using a Bayesian network (BN). The BN produced a single probability estimate calculated based on the prior and posterior probability of the CF and IvSRS. The results showed IvSRS performed better (p < 0.05) in classifying sleepiness to three states, compared to non-modified vSRS. Considering each CF and IvSRS as stand alone indicators, integrating all these information under a BN significantly improved the systems performance (p ≤ 0.05). In addition to being able to function well in the event of missing vSRS, the proposed system has a prediction horizon of 12 h, with F 1 -measure > 78%.