Emotion Detection from Physiological Markers Using Machine Learning

Human emotion recognition in computer and robotic systems is crucial because it allows these systems to respond to users in a way that feels natural and supportive. By interpreting emotional cues, these systems can adjust their interactions—offering empathy, encouragement, or even assistance during times of distress—enhancing user satisfaction and making technology more accessible and engaging. Emotion recognition methods include analyzing facial expressions, vocal tone, and physiological signals, with the latter being especially effective because physiological data offers objective, real-time insights that are less susceptible to misinterpretation or masking than visible expressions. In this paper, we conducted an experiment for emotion recognition from physiological markers using machine learning algorithms. Each of the participants involved in the experiment was exposed to video stimuli designed to evoke specific emotions. Using physiological markers such as heart rate (HR) and respiratory rate (RR), seven emotions—anger, sadness, fear, amusement, neutrality, surprise, and happiness/joy—were analyzed. Three classification methods Random-forest, SVM and J48 were used. According to the results from the experimental evaluation, the highest accuracy for classifying emotions, based on both HR and RR across all emotions, was obtained with J48 algorithm. Specifically, the most clearly expressed and distinguishable emotions through RR were fear and sadness, with classification accuracies of 96.43% and 92.86%, respectively. Sadness was also the most accurately classified emotion through HR, with an accuracy of 85.71%. Gender differences were noted, with females reacting more to sadness and males to happiness.