Efficient Analysis and Classification of Stages Using Single Channel of EEG Through Supervised Learning Techniques

Abstract

Sleep-related disorders have one of the challenging health issues across world. Identifying the sleep irregularities will be one important primary step of treatment for any types of sleep diseases. To perform day-to-day activities, proper healthy sleep is required for individual one’s life. This also plays one of the vital roles into the human life to maintain the proper health in both physically and mentally wise and which alternatively maintain our quality of life in smooth manner. The main objective of this research work is to propose a simple and efficient automated sleep stage classification methods based on single channel of electroencephalogram (EEG) signal using of machine learning techniques. Both time and frequency domain features are applied for that analysis of sleep quality and classifying the sleep stages for identification of sleep abnormality during sleep in night. Total 28 features are extracted from 750 epochs with 3000 sample points through C3-A2 channel of EEG signal. We obtained the sleep recordings from ISRUC-Sleep public sleep dataset, which is specifically designed for sleep study. The present research work is based on two-state sleep stage classification model through two machine learning classifiers such as support vector machine (SVM) and decision tree (DT) and used ten-cross-validation techniques and four evaluation parameters such as recall, specificity, precision and F1-score for measuring the performance of the proposed research work. The achieved results from evaluation matrices show an effective performance with SVM classifier. The overall accuracy achieved for two-state sleep classification problem is 95.60% through SVM and 91.20% through DT. Similarly for subject-2, the results achieved for SVM and DT are 87.46% and 87.06%, respectively. The proposed outcome shows that our two-state sleep classification system is similar and slightly better overall accuracy with compared to earlier publish similar methods research work. The corresponding Kappa coefficient scored for subject-16(1.92, 1.43) guarantees that excellent agreement and for subject-02 (0.72, 0.12) indicates that substantial agreement of the classification.