Ensemble learning versus deep learning for Hypoxia detection in CTG signal

Hypoxia is a condition of the decreasing oxygen supply on the fetal body tissues that will lead to fetal mortality. The experts will categorize fetal condition into two levels i.e. normal and hypoxia, based on CTG data analysis. Dataset which contain noises will affect to misinterpretation by the experts. The ensemble learning methods and deep learning methods are implemented to detect hypoxia. Ensemble learning models used include Bagging Tree, AdaBoost, and Vooting Classifier with classifier methods such as Decision Tree, SVM, SGD, GLVQ, and Naive Bayes. Deep learning models used include CNN and DenseNet. These methods are applied to CTG dataset, especially FHR signal. The classification processes utilize pH label as the benchmark. The benchmark is use to classify the dataset into two stage, normal and hypoxia. The best evaluation performance is obtained by Bagging Tree method with Naive Bayes Classifier. The F1-score for normal class was 0.76 and 0.45 for hypoxia class.