Heart Rate Variability Analysis through wearable devices: influence of artifact correction method on classification performance for emotion recognition

Heart Rate Variability (HRV) analysis is widely explored in several application fields, such as emotion recognition. Photoplethysmographic (PPG) signals are often considered for this analysis because of their large use in wearable devices. However, quality of these signals (in terms of added disturbances) could be not always optimal, since they are susceptible to many factors, e.g. motion artifacts, ambient light, pressure of contact, skin color and conditions. Therefore, methods for artifacts correction play a pivotal role and consequently influence the results. This paper aims at proposing a new data artifacts correction method to improve the classification performance in emotion recognition, considering PPG signals during audio stimulation, and a Support Vector Machine (SVM) classifier. Results show that the proposed method provides a better classification in stimuli detection (66.67%) with respect to data pre-processing performed with a standard tool (Kubios, 48.81%); however, for further improvement, other signals could be considered in combination with PPG, such as the electrodermal activity (EDA).