This study aimed to optimize emotion classification using electroencephalography (EEG) signals by evaluating the impact of window size, overlap, classification models, and wavelet transform selection. Understanding these factors is crucial for enhancing EEG‐based emotion recognition systems, which play a key role in improving human–computer interaction, adaptive user interfaces, and neurophysiological therapy. A comparative analysis was performed using supervised classification models with varying levels of window overlap and wavelet transforms to extract characteristics. The study followed four stages: (1) data preprocessing and wavelet selection for signal decomposition, (2) determination of optimal overlap and feature extraction, (3) application of supervised classification models, and (4) comparative evaluation using metrics such as the Area Under the Curve (AUC), accuracy (ACC), kappa coefficient, and F1‐score. The findings indicate that Random Forest (RF) and Support Vector Machines (SVM) achieve superior performance with overlap levels between 10% and 30%, regardless of the wavelet transform applied. In contrast, Logistic Regression (LR) and Decision Tree (DT) models exhibited lower accuracy and did not show significant improvements with varying overlap levels. Excessive overlap, however, degrades performance, emphasizing the importance of selecting an appropriate overlap level. Additionally, while wavelet transforms were used for feature extraction, the specific wavelet type (Coiflets, Daubechies, Symlet, Haar, or Meyer) did not significantly influence classification accuracy. These results highlight the importance of selecting appropriate overlap levels to optimize classification accuracy. This study provides valuable information for improving EEG‐based emotion recognition systems, which can be applied in fields such as brain–computer interfaces, mental health monitoring, and human–computer interaction. Future research should explore advanced feature extraction techniques and deep learning approaches to further improve classification performance.
Jarillo-Silva et al. (Thu,) studied this question.