Analysis of the Reduction of Diagnostic Models of the Eye Movement System in Psychophysiological Studies

This study investigates the reduction of informational models of the human eye movement system (EMS) constructed from experimental eye-tracking «input–output» data. Second-order Volterra integral models are employed to account for the dynamic and nonlinear properties of the system under investigation. Model identification is performed using the least squares method based on the EMS responses to test step signals. The resulting multidimensional transient characteristics are used to construct a set of diagnostic feature spaces, including a space of heuristic features, as well as spaces formed through sampling and wavelet decomposition. An analysis of model variability with respect to the respondent’s psychophysiological state is carried out, along with model reduction by selecting the most informative components. Based on the generated features, psychophysiological state classification is performed using a Bayesian classifier and the support vector machine (SVM) method. Classification performance is evaluated using the probability of correct recognition criterion, taking into account robustness to noise. The presented results confirm the feasibility of using quadratic models for constructing diagnostic features in intelligent technologies for psychophysiological state assessment.