Resumen
Obstructive sleep apnea (OSA), a common form of sleep apnea generally caused by a collapse of the upper respiratory airway, is associated with one of the leading causes of death in adults: hypertension, cardiovascular and cerebrovascular disease. In this paper, an algorithm for predicting obstructive sleep apnea episodes based on a spark-based support vector machine (SVM) is proposed. Wavelet decomposition and wavelet reshaping were used to denoise sleep apnea data, and cubic B-type interpolation wavelet transform was used to locate the QRS complex in OSA data. Twelve features were extracted, and SVM was used to predict OSA onset. Different configurations of SVM were compared with the regular, as well as Spark Big Data, frameworks. The results showed that Spark-based kernel SVM performs best, with an accuracy of 90.52% and specificity of 93.4%. Overall, Spark-SVM performed better than regular SVM, and polynomial SVM performed better than linear SVM, both for regular SVM and Spark-SVM.