Abstract
Intrapartum asphyxia is responsible for approximately 900 000 deaths per year worldwide. These numbers show the urgency of investing in the quality of fetal health care. The heart rate signal is a complex signal and sometimes behaves unpredictably. Thus, it becomes relevant to study approaches that take into account their complexity, namely non-linear compression-based methods. In this work, feature extraction was based on two approaches: univariate and bivariate. The univariate approach is concerned with the extraction of fetal, maternal and maternal-fetal compression ratios and the bivariate approach aims to extract compression indices from maternal-fetal heart rate simultaneous signals and of each of the signals individually over time. To understand how the features calculated in this work can be useful in distinguishing acidemic and non-acidemic cases, a classifier was applied. Three different classifiers were tested, and the one that proved to be more effective was the Support-Vector Machine. Furthermore, it was also possible to conclude that the input set of variables that provides a better performance (f1-score = 0.793) of the classifier is composed of the variables of maternal-fetal compression ratio, maternal-fetal normalized relative compression and maternal-fetal normalized compression distance, obtained through trend and residual signal, which indicates that slow and fast fluctuations on the heart rate time series are important in acidemia assessment.