An Improved Decision Support System for Automated Sleep Stages Classification Based on Dual Channels of EEG Signals

Abstract

An automatic analysis of sleep stage classification can reduce the use of manual annotating the sleep recordings from subjects during sleep time. This approach can alleviate the overburden of clinicians and this approach is indirectly helpful for sleep experts to take proper diagnosis of sleep-related diseases. For the proper treatment of sleep disorder the most important step is sleep scoring and computer-assisted sleep stage classification. With considering this above, two important steps our proposed research article is based on automated sleep stage classification in between two sleep classes such as wake versus sleep stages with high rates of accuracy and sensitivity. In this proposed study we have considered the EEG signals of one subject with suspected light sleep problem and the EEG signals of one healthy subject with prior no medication in subject to sleep problems. Here we have recorded the EEG signals from dual channels such as F3-A2 and C3-A2 and extracted epochs is 30 s. Eleven statistical features are extracted from each input channels. In this proposed work we have obtained the Online Streaming Feature Selection (OSFS) algorithm to identify the suitable features for classification task. The selected features classified by two conventional machine learning classifier such as DT and KNN. We have obtained tenfold cross-validation techniques to evaluate the system parameters. The result of our proposed experimental study is evaluated for two-state sleep classification sleep problems. The highest overall accuracy achieved for healthy subject through F3-A2 channel as 81.6% with DT classifier, and in same manner, the highest overall accuracy obtained for suspected light sleep problem subject through F3-A2 channel as 91.1% with DT classifier. The system performance toward sleep stage classification compared with the earlier contributed works mentioned in the state-of-the-art systems. The proposed research work can be used for health care application with huge PSG signal recordings for objective to improving in sleep stage classification system.