Understanding affective behaviour from physiological signals: Feature learning versus pattern mining

Monitoring emotions are gaining attention in the care of mental disease and behavioural health changes. To that end, there is an increasing interest in measuring emotions with sensors. In particular, deep learning approaches are being used for feature learning that enables emotion recognition. In this work, the focus is on using unsupervised techniques, as pattern mining, to characterise physiological signals from the classification of emotions with complementary predictive methods. An analysis is conducted to compare the performance results of the feature learning approaches regarding the pattern mining approach and the different approaches' properties.