Sleep Stage Analysis Based On Heart Rate Variability

Sleep is an important part of the body's self-repair.Good sleep can make people energetic,improve learning and working efficiency.Bad sleep can cause mental decline,malaise,accidents,even illness.The analysis of sleep quality can be used as a means of physiological health monitoring.Sleep stage classification is a relatively basic link in the analysis of sleep quality.Sleep stage classification based on heart rate variability compared with the traditional sleep stage classification based on EEG signals can overcome the limitations of the wearing of subject's sensor,the high experimental cost and others,making the development of portable sleep monitoring equipment possible.This article uses the relevant indicators of heart rate variability to build a sleep stage classification model,and the main work is as follows:Firstly,experimental data acquisition.The suitable data is selected as the data source by calculating the relevant data obtained from the MIT-BIH database in this thesis.The RR interval signals are extracted from experimental data sources and spline interpolation and re-sampling are performed to obtain RR interval signals with uniform time intervals.At the same time,sleep staging labels are extracted from experimental data sources.Secondly,the heart rate variability related features are extracted.Statistical methods are used to extract the time-domain characteristic RR interval mean MEAN,RR interval standard deviation SDNN,and the root mean square RMSSD of the difference between adjacent RR intervals;the wavelet transform,empirical mode decomposition and other methods are used to extracte the frequency-domain characteristics VLF,LF,HF,LF/HF,and TP;chaos and fractal methods are used to extract the nonlinear index box dimensions,and the Wolf method is used to extract the maximum Lyapunov exponent.The three types of the 10 kinds of heart rate variability features were combined into a feature vector,and then the feature vector is preprocessed and PCA dimension was reduced to reduce the data redundancy.Thirdly,the sleep stage classification model is set up.A one-to-one multi-class support vector machine solution is used in modeling,and a radial basis kernel function is selected as a kernel function of the support vector machine.At the same time,a combination of genetic algorithm and grid search algorithm is used to optimize the hyperparameters.The extracted time-frequency domain and nonlinear features are used as a feature vector.The SMO algorithm is used to train the sleep stage classification model.Fourthly,verify the experiment.First of all,the number of cross-validation k and the value of some parameter items ga_option in the genetic algorithm are selected through experiments.Second,the combination of genetic algorithm and grid search algorithm are used to optimize the hyperparameters,the best accuracy of the model corresponding to the optimization result is 80.5195% while the best accuracy of the model is obtained by using the grid search algorithm is 79.8701%.In comparison,the former has a better parameter optimization result.Then the support vector machine is trained,the trained model is used to classify the test data set,and the performance of the sleep stage classification model is evaluated using the confusion matrix.The evaluation results show that the average classification accuracy rate of the sleep stage classification model on the test set is 63.10%.The model has high classification accuracy which is as high as 93% for the LIGHT and DEEP stages,but it is not very good for two stages of WAKE and REM.At the same time,a model that does not include the two nonlinear features of box dimension and maximum Lyapunov exponent is designed as an experimental control group to evaluate its' performance.The experimental results show that the average classification accuracy rate of the model that do not include nonlinear features is 60.21%,which is lower than the accuracy rate of the model that contain nonlinear indicators.This result indicates that the classification accuracy of the sleep stage classification model using the two nonlinear features of box dimension and maximum Lyapunov exponent is more ideal.