Automated sleep scoring system using multi-channel data and machine learning

Sleep staging is one of the most important parts of sleep assessment and it has an important role in early diagnosis and intervention of sleep disorders. Manual sleep staging requires a specialist and time which can be affected by subjective factors. So that, automatic sleep-scoring method with high accuracy is beneficial. In this work 50 patients sleep data taken from 19 sensors of Philips Alice clinic polysomnography (PSG) device. There is an average of 4772801 data for each individual in a single channel, and approximately 87 million data is processed in 19 channels. Due to the large amount of data, after under sampling technique, dataset is created and Random Forest, Extra Trees and Decision Tree classifiers are applied on it. Although accuracy values vary from one person to another, average of 95.258% for Extra Trees, 95.17% for Random Forest and 91.318% for Decision Tree obtained. Furthermore, precision, recall and F1-score values were also 0.95362, 0.95258 and 0.94568 on average. Beyond the previous works in the area of sleep stage scoring, proposed work differentiated from them by having own database, providing higher accuracy and employing 19 channels. The results showed that the proposed work may alleviate the burden of sleep doctors and speed up sleep scoring. •Multi-channel data and machine learning are used for automatic scoring of sleep stages.•The proposed multi-layer pre-processing algorithm makes multi-channel learning possible.•Each sleep stage (WK, N1, N2, N3, and REM) is best represented using its distinctive features.•The method provides a better classification of sleep stages compared to existing approaches reported.