Algorithm 1 Process of proposed NGRMTE model for reproducibility.

Input—Raw physiological signals (EEG, EMG, EOG) Output—Detected sleep disorder class Pre-processing For each modality in (EEG, EMG, EOG)   Segment signals into 30-s epochs Feature Extraction For each 30-s segment in each modality   Extract time-domain features such as Mean, Standard deviation, Skewness, Kurtosis, and ZCR Extract frequency-domain features using FFT, such as Spectral centroid, Spectral entropy, and Spectral band power using Equations (1)–(6). Modality Embedding using 1D-CNN For each modality   Input extracted features into modality-specific 1D-CNN using Equation (7)   Obtain modality-specific embeddings (fixed dimension) Modality-wise Residual Gated Cross-Attention Fusion (MRGCAF) For each modality pair (EEG-EOG, EEG-EMG, EOG-EMG)   Compute cross-attention between modality embeddings   Apply the forget gate to filter redundant or noisy interactions using Equations (8) and (9)   Add a learnable residual path to preserve essential signal features   Fuse actual modality and filtered cross-modality features using Equation (10) Multi-Modal Transformer   Stack fused features from all modality pairs   Apply feed-forward networks   Positional encoding Classification   Flatten transformer output   Pass through linear layers with a learnable residual path using Equation (11)   Apply SoftMax activation for multi-class classification using Equation (12)   Use focal loss to address class imbalance using Equation (13)   Optimize using the Adam optimizer Return—Detected sleep disorder class for each epoch