The article presented by Kiruthiga et al studies optimized subject-independent emotion recognition (SIER) from Electroencephalography (EEG) signals using deep learning and optimization techniques. The proposed method, Optimized Node-Level Capsule Graph Neural Network (NLCGNN-SIER-EEG), utilizes the DEAP dataset, employing a Sparse Adaptive Bayesian Filter (SABF) for artifact removal and the New Generalized Fuzzy Transform (NGFT) for statistical feature extraction. A Node-Level Capsule Graph Neural Network (NCGNN) is then optimized using the Piranha Foraging Optimization Algorithm (PFOA) to classify emotions like calm, happy, sad, and angry. The optimized model achieved a final mean accuracy of 86.41% across all 32 folds using Leave-One-Subject-Out Cross-Validation (LOSO-CV), confirming its ability to generalize to individuals unseen during training. Compared to existing SIER methods, the proposed approach showed substantial enhancements, providing 81.32% high accuracy, 89.31% high recall, and 91.43% precision. The research successfully developed the optimized NLCGNN-SIER-EEG to detect SIER from EEG signals without the need for individual calibration, with the better performance metrics validating the model's high accuracy and generalizability for practical application.