Results and methodology for classifying high risk pilots using CANFLY: A cognitive health screening tool for aviators

• Published in: International journal of industrial ergonomics 2024-05, Vol.101, p.103578, Article 103578
• Main Authors: Van Benthem, Kathleen, Brightman, Kirsten, Riguero, Elizabeth, Herdman, Chris M.
• Format: Article
• Language: English
• Subjects: Aging; Aviation; Cognitive health; Machine learning; Online assessment; Virtual reality
• ISSN: 0169-8141; 1872-8219
• DOI: 10.1016/j.ergon.2024.103578

Cognitive health screening for aviators would assist in managing a shortage of experienced pilots. Extending pilot careers by optimizing their cognitive health would address both the number and quality of pilots available for airline and general aviation operations. The present work tested the validity of an online screening tool for pilots that measures aviation domain-relevant cognition. Sixty-five licensed pilots (18–80 years, M = 48.8, SD = 16.3) with varying levels of experience completed a 30-min online cognitive health screening tool for pilots. Risk status was determined via a novel metric using self-reported incidents. Machine learning algorithms identified the cognitive factors most useful in identifying pilots with increased risk for accidents and serious incidents. Support vector machines and boosted decision tree algorithms provided the most reliable and strongest classifications models of pilot risk. Findings support the use of this short online screening tool for highlighting performance issues with domain-relevant cognitive abilities based on the Dynamic Mental Model for pilots, such as situation awareness and prospective memory. Understanding personal cognitive challenges is the basis for customized skill maintenance designed to augment cognition for those interested in safely extending their piloting careers. •The CANFLY showed high precision and recall when classifying pilots at-risk of incurring a serious incident or accident.•Machine learning models resulted in low false positives and low false negatives.•Determining “ground truth” risk for the supervised machine learning models was achievable via self-report measures of incidents and accidents.•Aviation-specific cognition, e.g., situation awareness and prospective memory, were relevant features in machine learning models of pilot risk.•Self-monitoring of cognitive factors associated with risk of accidents may extend the flying careers of general aviation pilots.