Developing an Apnea/Hypopnea Diagnostic Model Using SVM

Among sleep-related disorders, Sleep apneahas been under more attention and it’s the most common respiratory disorder in which respirationceases frequently which can lead to serious health disorders and even mortality. Polysomnography is the standard method for diagnosing this disease at the moment which is costly and time-consuming. The present study aimed at analyzing vital signals to diagnose Sleep apneausing machine learning algorithms.

This analytical–descriptive was conducted on 50 patients (11 normal, 13 mild, 17 moderate and 9 severe patients) in the sleep clinic of Imam Khomeini hospital. Initially, data pre-processing was carried out in two steps(noise elimination and moving average algorithm). Next, using thesingular value decompositionmethod, 12 features were extracted for airflow. Finally, to classify data, SVM with quadratic, polynomialand RBF kernels were trained and tested.

After applying different kernel functions on SVM, the RBF kernel showed the most efficient performance.After 10 fold cross validation method for evaluation, the mean accuracy obtained for normal, apnea, and hypopnea modes were 92.74%, 91.70%, 93.26%.

The results show that in online applications or applications where the volume and time of calculations and at the same time the accuracy of the result is very important, The disease can be diagnosed with acceptable accuracy using machine learning algorithms.