Sleep States Classification Based on Single Channel Electrooculogram Signal Using Variational Mode Decomposition

To reduce the burden of the analysts for visual inspection of a big data size for sleep-scoring, an automatic sleep surveillance system is a prerequisite. Computer-aided sleep-scoring will also accelerate broad-ranging sleep study for the analysis of data. As most of the current sleep-staging projects are on the basis of multi-channel or several physiological signals which are not comfortable for the user, so automatic sleep-staging on one single channel system being trustworthy is still to be successful. For this work, a method based on single channel Electrooculogram (EOG) signal for computerized sleep scoring is proposed. In the suggested method, EOG signal epochs are decomposed into three modes using Variational Mode Decomposition (VMD) and multiple features like statistical measures, spectral entropy measures, RCMDE and Autoregressive modelling (AR) coefficients from different modes are extracted. Different classification models are examined for evaluating the results and Random-Forest-Classifier (RF) demonstrates most accurate result employing 10 fold cross-validtion. The efficacy of our system's algorithm against existing works in the literature shows that the suggested approach is similar to or show higher performance than previous existed methods. For the 6-states to 2-states sleep classification, the proposed algorithm provides 88.083%, 89.21 %, 90.57%, 93.05% and 96.537% overall accuracy respectively. In addition, the suggested algorithm for this work shows an accuracy of 65.092 % for the identification of sleep stage S1.