Emotion Recognition Based on XGBoost and HRV

Emotion recognition has a wide range of applications and already plays an important role in safe driving and health monitoring. Human are formed by the interaction of multiple complex physiological signals, so using artificial intelligence methods to accurately predict emotions is the main goal of the current field of emotion computing. There are two main categories of emotion computing approaches,one that uses surface signals from the body to predict emotion and the other that uses intrinsic physiological signals from the body to predict emotion. The combination of human emotional state and wearable device system generally uses non-invasive physiological signal acquisition,which can easily and quickly obtain the physiological parameter data and emotional state required by the user, and the user can get a better sense of experience. In order to improve the accuracy of emotion recognition,this paper uses quartile spacing to remove the outliers of physiological signals, XGBoost (Extreme Gradient Boosting) algorithm for emotion recognition is validated in the public dataset WESAD, and the feature fusion of acceleration, electrocardiography, electroencephalography, body temperature, myoelectricity, and respiratory frequency is used for emotion recognition,and the model is evaluated. The accuracy reached 97.88% and F1-score 96.45% for emotion recognition.