Reducing the number of EEG channels can significantly enhance classification accuracy and computational efficiency in emotion recognition tasks. However, optimizing channel selection remains challenging. This paper compares four approaches for channel selection in emotion classification: exhaustive, CSP (Common Spatial Pattern), PCA (Principal Component Analysis), and PSO (Particle Swarm Optimization). Using the DEAP, SEED, and MAHNOB-HCI datasets, which are widely recognized benchmarks for emotion recognition research, we evaluate each method across channel configurations from 1 to 32 channels. The exhaustive method serves as a baseline by evaluating all 32 channels for each participant. Our results show that PCA achieves optimal performance with 16 channels across all datasets, while PSO excels with just 2 channels, balancing accuracy and computational efficiency. CSP attains its highest accuracy with 8 channels but struggles with fewer channels, particularly in the MAHNOB-HCI dataset. This study underscores the trade-off between channel quantity and classification performance, highlighting that reducing channels can retain accuracy while improving processing speed. These findings provide valuable guidelines for designing efficient EEG-based emotion recognition systems across different datasets.