Multi Domain Feature Fusion from a combination of Empirical Wavelet Transform and Boruta feature selection for Improved Sleep Stage Classification

The potential of machine learning can be significant in revolutionizing clinical studies on human sleep. Enabling a machine to diagnose the sleep stages could alleviate the workload for clinicians. Here, we revisit the task of classifying sleep stages by applying machine learning algorithms. We have proposed a combination of Empirical wavelet transformation (EWT) based feature extraction and Boruta feature selection approach for automated classification of sleep stages. Various multi-domain features were extracted from intrinsic mode functions(IMF) of EWT. We have used two publicly available datasets in this research work. The proposed work performed better as compared to existing works in terms of classification accuracy.