Neural network for human cognitive state estimation

Abstract

A neural network (NN) has been applied to the human cognitive state estimation based on the set of physiological measures, heart rate, blood pressure, respiration rate, skin potential response (SPR), blink rate and saccadic eye motion rate have been chosen as the representative metrical indices reflecting human mental state. The qualitative tendencies of these measures have been taken as the inputs of the NN. The human cognitive states are categorized into six pre-specified states: (1) information acquisition (IA); (2) memory related (MR); (3) thought (TH); (4) motor action (MA); (5) emotion (EM); and (6) others (OT). The adopted network a is three layer feedforward network trained with a backpropagation algorithm with forgetting. Sets of training data for learning have been collected through laboratory experiments, in which the subjects were induced to undergo a specific sequence of cognitive states. The resultant NN showed superior capability of discriminating human cognitive states based on the pattern of the physiological measures.<>