Machine Learning for Sleep Scoring

Abstract: In today’s society, Artifical Intelligence (AI) has become one of the most controversial research topics. The Artifical Neural Network (ANN) is a form of AI designed to mimic the human brain and is typically used for pattern recognition. This is especially useful within the medical field where ANN:s have already been implemented in a variety of applications. This project focuses on the possibility of extending the use of ANN:s to also be used to facilitate diagnosing sleep disorders, by investigating how well they can classify sleep states in rats.Two different types of network architectures were considered in this project: one fed with information from one single sleep state, and one fed with information from multiple consecutive sleep states. The highest classification accuracy achieved by the networks were 96.78 % and 97.02 % respectively.The networks were fed with features extracted from the provided Electroencephalography (EEG) and Electromyography (EMG) data of the rats. This enabled a reduction of the complexity of the ANNs, which resulted in low training times. It was concluded that feeding the network with extracted features from the data was a good approach, and that a network with access to information about several sleep states classified with the best accuracy.