COA_DNN: a hybrid crayfish optimization with deep neural network for detection of rapid eye movement behaviour disorder

Rapid Eye Movement (REM) Sleep Behavior Disorder (RBD) is a serious condition often considered an early indicator of neurodegenerative diseases such as Parkinson’s Disease (PD) and Lewy Body Disease (LBD). Traditional diagnostic methods primarily rely on polysomnographic (PSG) assessments, which are both expensive and time-consuming, highlighting the need for automated and accurate detection techniques. This study introduces a novel deep learning framework, COA-DNN, which integrates the Crayfish Optimization Algorithm (COA) with a Deep Neural Network (DNN) to enhance feature selection and weight optimization. By reducing overfitting and improving feature representation from physiological sleep data, COA-DNN significantly enhances classification accuracy. Experimental results confirm that the proposed model achieves 99.8% accuracy, outperforming established deep learning models such as CNN (95.6%), LSTM (96.8%), Bi-LSTM (97.2%), CNN-LSTM Hybrid (98.1%), and Transformer-Based models (98.5%). Compared to these existing methods, COA-DNN delivers a 4.2% improvement in accuracy, along with notable gains in precision (5.1%), recall (3.1%), and F1-score (2.9%). Additionally, it surpasses traditional machine learning models such as Random Forest (94.5%), SVM (95.1%), and XGBoost (96.4%), demonstrating its robustness and superior generalization. The COA-driven feature selection process not only enhances classification performance but also reduces computational complexity by eliminating redundant features. A ROC-AUC score of 99.9% further validates the model’s effectiveness in accurately distinguishing RBD from normal and control cases. This research underscores the potential of combining bio-inspired optimization techniques with deep learning, paving the way for advanced automated diagnostic systems that can support the early detection and continuous monitoring of RBD.