Leveraging Multimodal and Feature Selection Approaches to Improve Sleep Apnea Classification Performance

Abstract

Obstructive sleep apnea (OSA) is a sleep disorder with long-term adverse effects such as cardiovascular diseases. However, clinical methods, such as polisomnograms, have high monitoring costs due to long waiting times and hence efficient computer-based methods are needed for diagnosing OSA. In this study, we propose a method based on feature selection of fused oxygen saturation and electrocardiogram signals for OSA classification. Specifically, we use Relieff feature selection algorithm to obtain robust features from both biological signals and design three classifiers, namely Naive Bayes (NB), k-nearest neighbors (kNN), and Support Vector Machine (DVM) to test these features. Our experimental results on the real clinical samples from the PhysioNet dataset show that the proposed multimodal and Relieff feature selection based method improves the average classification accuracy by 4.67% on all test scenarios.