A Sleep Stage Classification Algorithm Based On Radial Basis Function Networks

Based on the auto-regressive model power spectrum analysis of sleep signal in time-frequency domain, it is found that each sleep stage has its own unique power spectrum in each frequency band. The change of sleep phase is accompanied with the change of sleep signal spectrum. In this paper, we firstly study the original RBF neural network for automatic sleep staging and then propose an improved classification algorithm in which the power spectrum of each sleep stage known as frequency domain features and five another time domain features are calculated as input parameters. The proposed classification algorithm is tested on ISRUC-Sleep data set. Experimental results demonstrate that classification algorithm based on the improved radial basis function network is effective in accuracy and efficiency.