Marilen Benner

CHAPTER 5 114 D Peripheral blood Menstrual blood Figure 1. Machine learning approach using ensemble strategy to determine immune features that identify patients of recurrent pregnancy loss (RPL). (A) Schematic representation of study design. A flow cytometry based dataset was established to describe the immune profile of systemic or uterine immunity, through leucocyte isolation from peripheral blood (PB) or menstrual blood (MB), respectively. (B) Number of feature combinations yielding optimal classification accuracy. Eight machine learning classifiers were used to established a ranked list of top features that were recursively reduced. Accuracy is depicted as average and standard deviation of 10 runs. (C) Heatmap of the features allowing for optimal accuracy regarding data from PB (left panel) or MB (right panel). (D) Receiver operating characteristics (ROC) curve showing the average area under the curve (AUC) using PassiveAgressive classifier of PB (left panel) and MB derived immune features (right panel) after stratified 10-fold cross-validation. ROC curves of the remaining classifiers are depicted in Supplementary Figure S2.