Emotion Recognition Based on Galvanic Skin Response and Photoplethysmography Signals Using Artificial Intelligence Algorithms

Emotion recognition using a limited number of wearable non-invasive sensors and with low computational cost presents a significant challenge in contemporary research. In this study, we aim to investigate the feasibility of emotion recognition based solely on Galvanic Skin Response (GSR) and Photoplethysmography (PPG), employing a Bipartition Labeling Scheme. We trained and compared the performance of five shallow learning algorithms, namely K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Gradient Boosting Machine (GBM), and a Convolutional Neural Network with a Single Layer Perceptron Classifier (CNN-SLP) on two datasets, DEAP and K-EmoCon. This allowed us to compare two different paradigms: non-portable vs. wearable sensors. Our findings demonstrate that achieving accuracies of 0.7 or higher and f1-scores of 0.57 on valence and arousal is attainable with KNN, SVM, and GBM, suggesting the potential for utilizing these proposed biosignals in emotion recognition, provided enough data is gathered and a robust labeling scheme is employed. This challenge can be effectively tackled as wearables that facilitate extended data recording periods become more widespread.