Features extraction from human eye movements via echo state network

The paper develops a procedure for features extraction from eye movement’s time series aimed at age-related classification of humans. It exploits the properties of the echo state network (ESN) reservoir state achieved after its intrinsic plasticity tuning. A novel, recently proposed approach for ranking of dynamic data series using as single feature the length of the reservoir state vector reached after consecutive feeding of each time series into the ESN was investigated in details using eye tracker recordings of human eye movements during visual stimulation and decision-making process. Inclusion of other features like variance of ESN extracted feature for multiple similar stimulations as well as decision correctness allowed for better classification of test subjects. The results support the view that the metrics and dynamics of the eye movements depend little on age, though they are strongly related to the visual stimulation characteristics.