Electrocardiogram feature selection and performance improvement of sleep stages classification using grid search

Sleep analysis is often used to identify sleep-related human health. In many cases, sleep disorders could cause a particular disease. One of the approaches to detect sleep disorders is by investigating human sleep stages. However, the selection of the proper electrocardiogram (ECG) features is still considered challenging and becomes an issue to achieve the performance of the algorithm used. Therefore, it is necessary to investigate which ECG features are very significant to the performance of the algorithm. In this study, the support vector machine (SVM) method has been utilized to classify sleep stages into two classes namely awake and sleep. In order to improve the classification performances, an optimization method of grid search was used to find the best parameters of the SVM. Feature selection of information gain was then used to find the most significant ECG features. To validate the performance results, one leave-subject out cross-validation has been conducted during the implementation. There were ten subjects involved in this implementation. The ECG signals from those ten subjects were used to differentiate awake from sleep state. Based on the results, our method obtained an average accuracy of 85.46% a precision of 84.05% and a recall of 85.44% respectively.