Leveraging Machine Learning And Psychophysiology Data For Safer Skies

Abstract: Air safety continues to be a critical concern, frequently affected by human mistakes like pilot workload, fatigue, stress, and emotional fluctuations. Addressing these challenges, recent advancements have emphasized the integration of machine learning with psychophysiological data to better understand and monitor pilot behavior. This project presents a systematic review of current research in this domain, following a rigorous methodology to analyze 80 peer-reviewed studies selected from a total of 3,352 articles across five prominent electronic databases. The review focuses on key dimensions such as behavioral indicators, data types, preprocessing methods, machine learning models, and evaluation metrics. Findings suggest a predominant focus on cognitive factors like workload and fatigue, while emotional responses and attentional dynamics remain underexplored. Conventional preprocessing techniques and classical machine learning algorithms—particularly decision trees and support vector machines—are widely adopted, whereas the application of deep learning approaches remains limited. Critical research gaps have been identified, including a lack of emphasis on the impact of preprocessing strategies, challenges related to data imbalance, insufficient diversity in data collection environments, and minimal attention to model interpretability. This study highlights the need for future research to explore advanced preprocessing techniques, incorporate explainable AI, and capture a broader spectrum of behavioral data. Such directions are essential for building more robust, accurate, and human-centric safety systems in aviation. Keywords — Air safety, behavioral indicators, machine learning, pilot workload, psychophysiology data.