Abstract—Background: cardiotocography (CTG) has long been used in clinical decision making to help assess the fetus’ condition during pregnancy. However it’s usefulness in the detection of fetal acidosis is debated due to high inter and intraobserver variability and general difficulty in interpreting the signals. The introduction of automatic analysis methods aims to decrease these issues originating from human limitations, but additional questions still remain. There is no clear concession when is it most useful to perform CTG measurements and which time periods posses the highest predictive capabilities. Method: a database of 1932 patients was analyzed after baseline and feature extraction. Several machine learning methods (SVM,logistic regression, random forest, KNN) were compared based on accuracy, F1 score, recall, precision, sensitivity and specificity.Furthermore the database was divided, based on when the measurement was taken (relative to time of birth), and the accuracy of the methods was compared again at intervals of 1 to 24 hours. Results: from the machine learning methods the support vector machine using polynomial kernel achieved the highest scores(sensitivity of 55% and specificity of 56%). The inclusion of older measurements caused a decrease (≈20%) in the predictive performance of the models. Conclusion: the results show that in clinical decision making the most crucial fetal heart rate measurements are the ones that are taken the closest to birth.