Human Emotional State Estimation Evaluation using Heart Rate Variability and Activity Data

Abstract

Human-Robot Interaction (HRI) is one of the most rapidly emerging fields in robotic applications over the years. One direction of the improvements in the HRI field is by adding the capability of emotional understanding as a fundamental part of human-human interaction necessities. Human emotion understanding has been studied through the well-known Heart Rate Variability (HRV) analysis recently. In this paper, two different methods of classification are proposed to find the relations between activity, heart rate, and emotional states. Two individual k-Nearest Neighborhood (kNN)-based classifications used in the first method and implemented for each dataset of pre- processed accelerometer data and HRV data where both aim to estimate the user's emotion and activity data at the same time. The features of the frequency domain-based HRV data and the user's activity data are combined into a new dataset and two different classifiers of Multilayer Perceptron (MLP) and Support Vector Machines (SVM) were used in the experimental evaluations. Performance comparisons are presented to show the efficiency. Results from both methods are analyzed and reported in this paper.