Clustering-based Approach for Characterization of Patients with Preeclampsia using a Non-Redundant Feature Selection

Preeclampsia is one of the most frequent causes of maternal death during pregnancy. Knowing the characteristics of the patients can help to better distribute clinical care within the hospital. Thus, in this manuscript we present the application of clustering to characterize patients with preeclampsia. The dataset used in this work belongs to the patients with preeclampsia who were treated at the "IESS Los Ceibos" hospital in the city of Guayaquil. The feature subset selection (FSS) task has been considered for removing redundant features. However, according to some studies, the selection of non-redundant features is not assured when using only a filter or a wrapper FSS approach. As a result, we applied a methodology for training classification models that uses a combination of filter and wrapper FSS approaches to ensure that non-redundant attributes are selected during the data pre-processing phase. To carry out the clustering, the K-Means and EM algorithms were chosen. Based on the comparison of the 6 applied models, the three-cluster model with K-Means was chosen since it is the one that shows the greatest differences between them.