Comparison of SFS and mRMR for oximetry feature selection in obstructive sleep apnea detection

Abstract

Obstructive sleep apnea is a disorder characterized by pauses in respiration during sleep. Due to this disturbance in breathing, there is a decrease in the oxygen saturation (SpO2) level. Thus, SpO2 can be used as a source of information for the automatic detection of apnea. Several solutions exist in the literature where different features are used. To find a better discriminant capacity, a subset of few features that obtains higher accuracy with the proper classifier is needed. To face this challenge, this work compares two different feature selection methods. The first one is a filter method named minimum redundancy maximum relevance, and the other one is called sequential forward search. These methods are tested with different classifiers. Two public datasets with 8 and 25 subjects are used to test and compare the performances of the different feature selection methods. A set of features for each classifier is obtained, and the results are compared with the previous work. The results found in this work show a good performance with respect to the state of the art and present a good option for apnea screening with low resources.