Highly imbalanced datasets occur in domains like fraud detection, fraud prediction, and clinical diagnosis of rare diseases, among others. These datasets are characterized by the
existence of a prevalent class (e.g. legitimate sellers) while the other is relatively rare (e.g. fraudsters).
Although small in proportion, the observations belonging to the minority class can be of a
crucial importance. In this work we extend an unsupervised learning technique-Self-Organizing
Maps-to use labeled data for binary classification under a constraint on the proportion of false
positives. The resulting technique was applied to two highly imbalanced real datasets, achieving
good results while being easier to interpret.