Classification Algorithms in Sleep Detection - A Comparative Study

This paper presents a comparison of different machine learning algorithms applied to automatic sleep detection which uses electroencephalogram signals as a differentiating basis. The Single-Layer Perceptron, Multi-Layer Perceptron, Support Vector Machine, Boosted Tree and the Multi-Agent (comprising of the earlier models) models are developed and analyzed with training and testing datasets. The results of the models are evaluated using a cross-validation technique. The models are compared with each other using the Cohen’s index, the True Positive Rate and True Negative Rate. The models are very successful with sleep stage detection reaching up to 94 %, and Cohen’s index reaching up to 0.69, showing considerable promise for deployment and future studies.