Preliminary classification of cognitive load states in a human machine interaction scenario

In this work, different cognitive load situations are examined and classified in the context of a Human Computer Interaction (HCI) scenario. Machine learning methods were used to detect three cognitive load states (overload, underload, normal load) with the help of five different psychophysiological signals (ECG, EMG, Respiration, GSR, Temperature). At first it is shown, that the three regarded states can be clearly distinguished in the Valence-Arousal-Dominance space (VAD). After this comparisons between a 10-fold-valdidation and a batch-validation as well as three different classifiers (k-Nearest-Neighbour, Naive Bayes, Random Forest) are accomplished. At last the influence of gender in contrast to an overall analysis is shown.