Detecting pilots’ expertise using transition matrix measures: a machine learning approach

Abstract

Visual perceptual skills are considered to be a crucial ability accounting for the advantage of highly trained experts in many domains (Li et al., 2012). Indeed, expertise exerts a top-down modulation on gaze behavior and strategies. In this sense, experts with extensive training, domain knowledge, and experience can perceive important relationships among multiple information, enabling them to orient their attention toward relevant information and identify abnormalities with a high efficiency (Hoffman and Fiore 2007; Palmeri et al. 2004). Multiple studies investigated differences in scan paths and scan patterns between novices and experts from different domains (Law et al, 2004; Ooms et al., 2014). In aviation, the literature also emphasizes different visual scanning strategies in novices vs expert’s pilots (Kasarskis et al., 2001; Yang et al., 2013). In particular, early work investigated the effects of expertise on the visual scanning of flight instrument (Fitts et al., 1949). Later, Tole et al. (1983), and more recently, machine learning approach was applied to this type of topic. For example, Hayashi (2005) proposed a Hidden Markov Model (HMM) approach based on gaze behavior in a space shuttle crew: with Hidden Markov state corresponding to different flight tasks.