Clustering of emotional states under different task difficulty levels for the robot-assisted rehabilitation system-RehabRoby

In this paper, we study an unsupervised learning problem where the aim is to cluster the emotional state (excitedness, boredom, or stress) using the biofeedback sensor data while subjects perform tasks under different difficulty levels on the robot assisted rehabilitation system-RehabRoby. The dimension of the training vectors has been reduced by using the Principal Component Analysis (PCA) algorithm after collecting the biofeedback sensor measurements from different subjects under different task difficulty levels to better visualize the sensor data. The reduced dimension vectors are fed into a K-means clustering algorithm. Numerical results have been given to demonstrate that for each training vector, the emotional state decided by the clustering algorithm is consistent with the subjects declaration of his/her emotional state obtained via surveys after performing the task.