Development of Real-Time Electroencephalography-Based Emotion Recognition Models Using Machine Learning Techniques

Emotion recognition using electroencephalography (EEG) signals has gained significant attention in affective computing, human-computer interaction, and mental health applications. Owing to the complex and non-stationary nature of EEG signals, diverse feature extraction methods - spanning time-domain (statistical), frequency-domain, and deep learning approaches - have been explored to capture intrinsic signal characteristics. This study presents a critical evaluation of three machine learning models for EEG-based emotion classification, leveraging both time- and frequency-domain features. Specifically, the performance of eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and K-Nearest Neighbors (KNN) was investigated using the DEAP (Database for Emotion Analysis using Physiological Signals) dataset, a widely adopted benchmark for emotion recognition research. To enhance signal quality, EEG data were segmented into 1-second epochs, from which a comprehensive set of statistical and mathematical descriptors were extracted: mean, standard deviation, root mean square, minimum, maximum, peak value, activity, mobility, complexity, entropy, skewness, and kurtosis in the time-domain, along with entropy in the frequency-domain. A 10-fold cross-validation procedure was applied under a cross-subject evaluation scheme to assess generalization across individuals. Experimental results demonstrated that KNN achieved the highest classification accuracy (96.77%), surpassing XGBoost (88.55%) and LightGBM (84.08%), thereby challenging the conventional preference for more complex ensemble models in EEG-based emotion recognition. These findings highlight the critical role of feature engineering and classifier selection in optimizing system performance. The outcomes have practical implications in clinical diagnostics (e.g., early detection of depression and anxiety) and adaptive systems, including mood-aware interfaces, neurofeedback applications, and emotionally adaptive gaming environments.