Abstract
Known as a pregnancy complication due to high blood pressure and may be accompanied
by damage to another organ system, preeclampsia afflicts between 3 and 6 percent of US
pregnancies each year. Studies have shown the importance of early detection of preeclampsia
to prevent further complications that are detrimental to both mother and infant. In this
work, we develop an algorithmic modification of Deep Neural Networks to identify high risk
patients in preeclampsia diagnosis using imbalanced datasets in the presence of missing
values. We identify the most influential set of clinical features relevant to preeclampsia and
train a classifier that can be embedded within a clinical decision support system. Our results
provide evidence in favor of increased consideration of patient race/ethnicity in preeclampsia
prediction, and for more personalized medicine in general.