<p>Suboptimal spatial aggregation of neural signals often constrains the efficacy of electroencephalography (EEG) for emotion recognition. Although functional connectivity measures are increasingly employed, the critical influence of region of interest (ROI) configuration on classification performance remains largely unquantified. This study presents a systematic evaluation of how varying ROI definitions affect EEG-based emotion recognition accuracy. We designed five distinct, ROI configurations (CROI A-E) to group EEG electrodes, moving beyond single-atlas paradigms to probe the effects of hemispheric division and functional segregation. Phase-locking and mutual information were extracted from each configuration to capture linear and non-linear functional connectivity. These features were used to classify the four quadrants of the valence-arousal space within the DEAP dataset using a support vector machine and k-nearest neighbor classifiers. Our results demonstrate that the spatial architecture of ROIs is a principal determinant of model performance, with a specific configuration achieving a maximum classification accuracy of 92.65%. This finding establishes that the strategic definition of ROIs is not merely a pre-processing step but a critical hyperparameter in the design of neural affective computing systems. This study provides a rigorous, empirical framework for optimizing spatial analysis in EEG-based emotion recognition.</p>