Abstract:
The method that heart rate variability signals automatic detecting sleep apnea is proposed. The method includes three stages. First, extract heart rate variability signals from ECG signals, and perform outlier processing and segmentation. Secondly, the two-stage feature selection strategy based on analysis of variance (ANOVA) and max-relevance and min-redundancy (mRMR) proposed in this thesis is used to obtain feature vectors. Finally, a random forest classifier is trained using 5-fold cross-validation. The algorithm uses 9 features to classify each 1-minute “epoch” as apnea or normal. The average accuracy of sleep apnea recognition obtained in the test set is 80%, and the Kappa coefficient is 0.61. This method has the characteristics of non-invasive and low cost, which is beneficial to the hardware implementation of portable sleep detection devices.
