<p>Obstructive sleep apnea (OSA) is a common yet underrecognized sleep disorder resulting in major cardiovascular and metabolic morbidity. Polysomnography, which is the gold standard for diagnosis, is accurate but not convenient or scalable for early screening, as it is resource-intensive. Background This study analyzes the feasibility of using clinical questionnaire-based data in machine learning (ML) models to predict OSA risk. Data on demographic, anthropometric, and symptom-related variables, collected from standardized sleep-related clinical questionnaires, were preprocessed and with them, multiple ML algorithms (Logistic Regression, Random Forest, Gradient Boosting and Support Vector Machines) were trained and evaluated. The accuracy, sensitivity, specificity, F1-score, and Area Under the Curve (AUC) of the Receiver Operating Characteristic for AUC-ROC (Receiver Operating Characteristic). Through comparison, ensemble-based techniques, especially Gradient Boosting, yielded better outcomes than other models with AUC-ROCs of 0.89 along with high sensitivity for identifying moderate-to-severe OSA cases. Analysis of feature importance revealed that the most predictive variables were BMI, neck circumference, Epworth daytime sleepiness score and snoring frequency. Conclusions these findings indicate that questionnaires-based ML models can be cost-effective, non-invasive and scalable tools for early risk assessment that may address diagnostic delays and triage patients optimally for confirmatory testing. Optimal screening for obstructive sleep apnea (OSA) with clinical questionnaires at the level of primary care provides a significant opportunity for early intervention and, potentially, preventive health.</p>