Drowsy driving monitorization using statistical and machine learning techniques

Sleep is crucial to people’s health and well-being, that is essential for cognitive function, emotional regulation, and physical health. Therefore, the quality of sleep directly influences daily performance, influencing our emotional stability and memory consolidation to decision making. Despite its importance, inadequate sleep contributes to an array of health problems and diminished quality of life. Consequently, sleep-related issues impact not only driver safety but also that of passengers, pedestrians, and other road users. This thesis seeks to comprehend sleep disorders among Portuguese drivers, with the objective of addressing prevailing knowledge gaps across all districts. A questionnaire-based approach, covering sleep disorders provided valuable insights. Among the findings, a significant portion of drivers reported poor sleep quality (60.2%) with a subset experiencing excessive daytime sleepiness (38.8%). Additionally, the study addresses the alignment between circadian rhythms and work schedules, acknowledging its impact on productivity. The analysis reveals that while the majority of drivers have work schedules aligned with their natural circadian rhythms, those whose schedules diverge from this alignment tend to experience increased daytime sleepiness. Furthermore, the study aims to improve road safety by creating affordable solutions that seamlessly integrate into driving routines, with a specific focus on combating drowsiness while driving. Driving simulations were conducted, and data were collected using a wearable device (Empatica E4), from which heart rate variability data was acquired. This data was then utilized for the classification and prediction of drowsiness. The study addressed challenges associated with subjective drowsiness classification (awake or drowsy) and used multivariate statistical process control techniques to improve the reliability of these classifications. Therefore, a comprehensive analysis revealed promising results, with the Ensemble Tree (ET) model emerging as the best classifier. Additionally, regression models, particularly the XGBoost (XGB), exhibited the ability to predict drowsiness with a lead time of two minutes, outperforming the ET model. Notably, the superior volume of data on drowsy events raised concerns regarding the model’s ability to recognize wakefulness accurately. After adding new data, the model was able to correctly distinguish between those who are awake and those who are sleepy. These findings are significant and promising in terms of their potential to predict drowsiness in advance.

Name thesis:degree_name

Programa doutoral em Industrial and Systems Engineering

Grantor

Universidade do Minho

Year dc:date.issued

2024

Author dc:creator

• Antunes, Ana Rita Oliveira

Advisors dc:contributor.advisor

• Braga, A. C.
• Gonçalves, Joaquim José de Almeida Soares

• Drowsiness at the wheel
• Feature selection
• Machine learning
• Sleep disorders
• Statistic
• Aprendizagem automática
• Distúrbios de sono
• Estatística
• Seleção das variáveis
• Sonolência ao volante

Statement dc:rights

• openAccess

Licence dc:rights.uri

• http://creativecommons.org/licenses/by-nc-sa/4.0/

Language dc:language.iso

eng

Handle dc:identifier.uri

https://hdl.handle.net/1822/92959

OAI record oai:identifier

http://repositorium.sdum.uminho.pt/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Arepositorium.uminho.pt%3A1822%2F92959