Diagnosis model for obstructive sleep apnea combining artificial immune system and logistic regression: A case study
1Department of Industrial Engineering and Management, Ming Chi University of Technology, New Taipei City, Taiwan.
Objective:
Obstructive Sleep Apnea (OSA) is a common sleep-related breathing disorder associated with substantial global health and economic burdens. Although polysomnography (PSG) is the diagnostic gold standard, its high cost and limited accessibility constrain its large-scale clinical application. This study aimed to develop and validate an artificial intelligence-based OSA diagnostic model, termed AISLR, which integrates an Artificial Immune System (AIS) for feature selection with Logistic Regression (LR) for classification, thereby enabling accurate and efficient OSA risk prediction using home sleep testing (HST) data.
Methods:
A total of 3,345 trainees (2,625 men and 720 women) participated in this study. The dataset was randomly split into training and testing subsets (90:10) to create an internal validation framework. Model training was performed using 10-fold cross-validation within the training set, and the performance was evaluated on the held-out test set. All experiments were repeated 30 times using different random seeds to ensure robustness. The AISLR was compared with classical supervised learning models, and its performance was assessed using accuracy, AUC, Matthews correlation coefficient, and geometric mean to account for class imbalance.
Results:
The AISLR model achieved the highest classification performance (G-mean = 0.9960). Supervised feature selection identified five key HST-derived indicators related to respiratory events and arousals. Leave-one-feature-out validation using SHAP demonstrated strong ranking consistency with stable ordering and substantial overlap, while dose-response analysis revealed monotonic risk trends across feature quantiles. Comparison with an unsupervised HVGS baseline showed agreement with intrinsically high-variance features, supporting the robustness and structural stability of the identified indicators beyond label-driven optimization.
Conclusion:
The proposed AISLR model provides an accurate, cost-effective, and scalable approach for OSA risk assessment using readily accessible HST-derived parameters, thereby reducing dependence on PSG. This framework has the potential to enhance early screening, community-based diagnosis, and telemedicine applications for sleep-disordered breathing.
