A Comparative Analysis of Regression Algorithms for Prediction of Emotional States using Peripheral Physiological Signals

Human emotions play a key role in making decisions and solving problems in various situations in our daily lives. Poor mental health can reduce our body's immune power paving the way to mental health issues like depression, anxiety and stress. A healthy amount of stress may be motivating and useful, but excessive, continuing stress or a strong reaction to stress can cause obesity, heart attacks, type 2 diabetes, stomach pains, body aches, menstrual cycle disruption, meningitis, and other health problems. Due to this reason, it has become essential to perform periodic research on both the identification and prediction of emotional states. Wearables and other Artificial Intelligence (AI) based technologies for monitoring mental health might totally replace or supplement psychologists as technology advances. This study aims to predict valence and arousal in emotional dimensions space using different physiological biomarkers collected in K-EmoCon dataset through wearable sensors and compare the outcomes of several prediction models using supervised machine learning techniques. Two emotional dimensions were considered namely valence and arousal in predicting emotions on a continuous scale using various regression algorithms like Decision tree, Random Forest, Multi-Linear, polynomial, XG boost, Ridge, and lasso model. These models are used to find the relevance of biomarkers and their contribution to predict different emotional states through different metrics like R
²
score, Mean square error (MSE), Mean absolute error (MAE), and Root mean squared error (RMSE). According to experimental outcomes, Random Forest regressor outperforms the other algorithms with an R
²
score of 0.84 and MSE of 0.27 respectively.