Detection of fatigue by machine learning methods based on human physiological data

Abstract

Fatigue is a prevalent issue in today’s society, affecting individuals across all age groups and professions. The consequences of fatigue are particularly critical in professions where human performance directly impacts safety and efficiency, such as in healthcare, transportation, aviation, emergency services or military operations. Fatigue, if undetected, can lead to significant risks, including accidents and decreased productivity. Previous research has shown that applying machine learning techniques to eye-tracking data can assist in detecting fatigue. However, existing methods have demonstrated only limited accuracy. The primary goal of this thesis is to enhance the classification accuracy of fatigue detection. This study utilizes eye-tracking data from a study conducted by the German Aerospace Center, involving 66 healthy young participants (34 males and 32 females) with a mean age of 25.5 ± 4.6 years, recorded in a control-center-like environment. Machine learning algorithms, combined with eye-tracking technology, offer an automatic, and non-intrusive method for fatigue detection by measuring eye movements that are potentially associated with fatigue. To improve accuracy, this thesis introduces the creation of Areas of Interest (AOIs) and integrates gaze behaviour analysis into the conventional eye-tracking data analysis. New features were developed and combined with existing eye-tracking features to enhance predictive performance. The results of this thesis indicate that fatigue can be more accurately classified when AOIbased gaze analysis is included. However, there remains potential for further improvements in fatigue classification through the use of advanced neural networks, such as Convolutional Neural Networks (CNNs), which could provide even greater accuracy.