Decoding the Night: A Machine Learning Approach to Predict the Severity of Obstructive Sleep Apnea through Clinical Parameters

<bold>Introduction:</bold> The definitive diagnosis of obstructive sleep apnea (OSA), a type of sleep disorder, is made through polysomnography. Machine learning methods can be utilized in polysomnography reporting and diagnostic procedures. The aim of this study is to evaluate the possibility of determining the severity of OSA using machine learning methods. <bold>Methods:</bold> In this study, an artificial neural network (ANN) model was developed for the detection of OSA. The dataset used for training and testing comprised 25 features (Figure 1) collected from 339 patients (67% for training, 33% for testing). <bold>Results:</bold> During training, the highest accuracy value of 87.67% was achieved with the training dataset, while the highest accuracy value of 80.35% was obtained with the test data. To demonstrate the validity of the model, a ROC curve has been shared in Figure 2. <bold>Conclusions:</bold> The accuracy values obtained on the test data also demonstrate the model's generalization ability. ROC curve analysis indicates that the model provides high sensitivity and specificity in detecting OSA. In future studies, utilizing a larger patient sample and incorporating different features could further enhance the model's performance.