Comparison of Machine Learning Based Emotion Recognition Models Trained Using Physiological Signals

In this study, Blood Volume Pulse (BVP), Electrodermal Activity (EDA), and Skin Temperature (ST) data were collected from 15 adult participants to be used in robot-assisted rehabilitation studies. The collected data were classified into two categories based on the combination of three emotional states: Positive (P), Negative (U), and Neutral (N), using machine learning methods. For classification, various machine learning models such as Cascaded Random Forests, Support Vector Machines, K-Means, K-Nearest Neighbor, and XGBoost. The study results showed that the Cascaded Random Forest model achieved the highest accuracy rate (77% and 76%) for the PU and NP classes, respectively, while the K-Means model achieved the highest accuracy rate (75%) for the NU class. Additionally, it was found that the accuracy rate was higher when standardization was not applied to the collected physiological data.