Unsupervised Assisted Sleep staging Classification Algorithm under Fuzzy Few Samples

Sleep staging has a strong reference value in modern medicine for doctors to judge patients’ physical and mental state and provide treatment advice. However, in reality, according to the original information of sleep Electroencephalogram (EEG), it is difficult for doctors to manually judge, and sleep staging samples are difficult to obtain, so the data is few. At the same time, the robustness of the sleep staging model obtained only by individual learning is poor. In order to solve the problem of using fuzzy few samples to design the sleep staging prediction model to provide accurate sleep staging information for doctors, an unsupervised auxiliary algorithm model is designed. Firstly, according to the data characteristics of sleep EEG signals, low-pass filtering and fast Fourier transform were performed on the EEG signals recorded during sleep. Sleep stages are performed according to the frequency parameters, and normalization is performed to highlight the wave characteristics of different components. Secondly, due to the existence of different sample data in each stage, unsupervised samples are classified and corrected by K-Means clustering method, and a more robust model is trained under the premise of ensuring the diversity of training samples. Finally, the data set divided by clustering is sent to Support Vector Machine (SVM) classification learning, and the Gaussian kernel function is used to achieve high-dimensional mapping, which can reduce the impact of deviation from the center data on the sample center. The sleep staging classification algorithm designed in this paper can classify the sleep staging under the condition of fuzzy few samples, in the case of equal proportion of training set and test set, the correct rate is higher than 90 %, and in very few samples, the classification accuracy is more than 85 %.