Identification of sleep arousals from physiological signals using Machine Learning classification algorithms
Vinodhini Ranganathan, Abijeet Waghmare, Uma Maheswari Krishnaswamy, Tony Raj
St. John’s Research Institute, Bangalore
Aims: This paper attempts to correctly classify arousal regions of individuals from their sleep study data. This sleep study data was collected from five different physiological signal recordings namely electroencephalography, electrocardiography, electromyography, electrooculography and oxygen saturation. The arousal regions have already been annotated and the goal of the challenge is to achieve high accuracy of classification using various computational algorithms.
Methods: The given training data of 989 individuals has already been annotated. We have used six different machine learning algorithms namely Extreme Gradient Boost (XGB) classifier, Support Vector Machine (SVM), Logistic Regression, Decision Tree, ADAboost & Gradient Boost to classify the target arousal regions on the training data set.
Results: Amongst the 1983 patients included in this study, (994-training, 989-test) the algorithms were run on the training dataset. Cross validation was done on the training dataset to get validation metric which consisted of accuracy, precision, recall and F1 score. Highest accuracy of 64.19% was achieved with the XGB classifier (Precision 0.5, Recall 0.75, F1 Score 0.6). Results for other models are as listed below. Logistic Regression (54.85, 0.44, 0.88, 0.58); Decision Tree (52.89, 0.45, 0.63, 0.53); Adaboost Classifier (61.18, 0.42, 0.63, 0.50) gradient boost (58.97, 0.44, 0.50, 0.47) SVM classifier has accuracy of 62.8