Applying machine learning approaches to assess cardiotocography exams

The fetal cardiac frequency is accepted as a reliable parameter to determine the fetal welfare. Cardiotocography is the method which allows studying this frequency and its variations as a response to the uterine activity, allowing the interpretation of the different fetal behavioral states. The aim of this work is to assess the performance of different machine learning algorithms for cardiotocography exams classification. Two classification tasks are considered: the first corresponds to characterizing the exam according to the fetal state; the second corresponds to classifying the exam according to the morphological pattern. Our experimental results indicate that the tested algorithms have a promising performance for both tasks. Among the checked algorithms, Random Forest has the best results, achieving an accuracy of 94,9% for the fetal state classification task and an accuracy of 87,3% for the morphological pattern task.