Single-Channel vs Dual-Channel EEG Analysis for Sleep Stage Detection

The optimal electrode configuration for sleep monitoring remains an important question for practical applications. This study investigates single versus dual-channel approaches for sleep stage classification using Fpz-CZ and Pz-Oz recordings from the Physionet dataset. We develop optimized models for each channel independently and explore multiple fusion strategies including early, late, and intermediate fusion with attention mechanisms. Results demonstrate that dual-channel approaches achieve 92.4% accuracy, outperforming single-channel methods by 4.8% for Fpz-CZ and 7.2% for Pz-Oz. However, channel contribution analysis reveals sleep-stage dependent patterns: Fpz-CZ better captures slow-wave activity in deep sleep, while Pz-Oz excels at detecting alpha rhythm during wake and REM periods. Our reduced-channel transfer techniques maintain 96.3% of dual-channel performance when only one electrode is available. The adaptive channel selection mechanism further improves robustness by dynamically switching channels based on signal quality. These findings provide critical insights for electrode placement optimization in practical sleep monitoring applications, enabling more targeted and efficient EEG recording configurations.