XGBoost-Based Emotion Classification in Healthcare Using Electroencephalogram Signals

Over the past few years, significant research has been done on the brain signals employed to detect emotions. Researchers have implemented machine-learning techniques to identify emotions as psychological variables, but attaining exact results remains challenging. This paper aims to improve the ability to accurately classify an individual's affective states using electroencephalogram (EEG) signals. The process analyzes the power spectral density between brain rhythms to extract significant features from EEG signals. Max-depth and n-estimator selection are utilised to improve the feature extraction process. Then, we utilise the Extreme Gradient Boosting (XGBoost) to classify the data. The Gradient boosting framework is the basis of XGBoost, an ensemble of linear model solutions with tree-based learning methods to improve predictive accuracy. The proposed approach was evaluated against various classification techniques on the DEAP dataset. These techniques included feedforward neural networks (FNN), AdaBoost, K-Nearest Neighbours (KNN), and Random Forest. The emotions that were modeled were dominance, arousal, liking, and valence. The proposed method was more precise than other techniques in classifying emotions: 86.39% for liking, 83.60% for arousal, 84.09% for dominance, and 84.87% for valence. The results indicate that the proposed method performs well at enhancing the precision of mood classification and has the potential to be applied in affective computing and brain-computer interface (BCI) systems.