Browsing by Author "Alreshidi, Ibrahim M."

Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Flight crew’s cognitive states detection using psychophysiological measurements and machine learning techniques
    (Cranfield University, 2023-10) Alreshidi, Ibrahim M.; Moulitsas, Irene; Jenkins, Karl W.
    In the ever-evolving landscape of aviation safety, the accurate assessment of pilots' mental states is of paramount significance. This thesis elucidates the critical role of Electroencephalogram (EEG) data in comprehending pilots' cognitive conditions. The dataset, sourced from attention-related human performance limiting states, was publicly available on the NASA open portal website and encompasses EEG, electrocardiogram, galvanic skin response, and respiration data. The initial analyses delved into the challenges posed by noise within EEG recordings. After rigorous testing, it was observed that prevalent preprocessing techniques, specifically band-pass filtering coupled with Independent Component Analysis, were not always effective. This inefficiency underscored the need for more advanced methodologies to optimize machine learning outcomes. In response, subsequent research stages proposed a hybrid ensemble learning approach. This innovative approach integrated advanced automated EEG preprocessing with Riemannian geometry. Through rigorous experimentation and validation, it was determined that this methodology accentuated the profound advantages of refined preprocessing, significantly enhancing the accuracy and reliability of EEG data interpretation. As the inquiry advanced, a more integrative approach was adopted, amalgamating EEG with other physiological data. A novel methodology, synergizing one-dimensional Convolutional Neural Networks with Long Short- Term Memory architectures, was unveiled. Additionally, the impact of employing methods to handle data imbalance on machine learning performance was thoroughly examined. In the concluding phases, the research placed a heightened emphasis on model interpretability. Through the integration of SHapley Additive exPlanations values, a bridge was constructed between intricate model predictions and nuanced human comprehension, delineating paramount features for distinct cognitive states. To encapsulate, this thesis offers a meticulous dissection of EEG data manipulation, machine learning, and deep learning constructs, positing a blueprint for the augmentation of aviation safety through in-depth cognitive state evaluations.
  • Thumbnail Image
    ItemOpen Access
    Miscellaneous EEG preprocessing and machine learning for pilots' mental states classification: implications
    (2022-12-12) Alreshidi, Ibrahim M.; Moulitsas, Irene; Jenkins, Karl W.
    Higher cognitive process efforts may result in mental exhaustion, poor performance, and long-term health issues. An EEG-based methods for detecting a pilot's mental state have recently been created utilizing machine learning algorithms. EEG signals include a significant noise component, and these approaches either ignore this or use a random mix of preprocessing techniques to reduce noise. In the absence of uniform preprocessing procedures for cleaning, it would be impossible to compare the efficacy of machine learning models across research, even if they employ data obtained from the same experiment. In this study, we intend to evaluate how preprocessing approaches affect the performance of machine learning models. To do this, we concentrated on fundamental preprocessing techniques, such as a band-pass filter and independent component analysis. Using a publicly accessible actual physiological dataset gathered from a pilot who was exposed to a variety of mental events, we explore the influence of these preprocessing strategies on two machine learning models, SVMs and ANNs. Our findings indicate that the performance of the models is unaffected by preprocessing techniques. Moreover, our findings indicate that the models were able to anticipate the mental states from merged data collected in two environments. These findings demonstrate the necessity for a standardized methodological framework for the application of machine learning models to EEG inputs.