Resumen de Distance Metric Learning with Prototype Selection for Imbalanced Classification
Juan Luis Suárez Díaz, Salvador García López 
, Francisco Herrera Triguero
Distance metric learning is a discipline that has recently become popular, due to its ability to significantly improve similarity based learning methods, such as the nearest neighbors classifier. Most proposals related to this topic focus on standard supervised learning and weak-supervised learning problems. In this paper, we propose a distance metric learning method to handle imbalanced classification via prototype selection. Our method, which we have called condensed neighborhood components analysis (CNCA), is an improvement of the classic neighborhood components analysis, to which foundations of the condensed nearest neighbors undersampling method are added. We show how to implement this algorithm, and provide a Python implementation. We have also evaluated its performance over imbalanced classification problems, resulting in very good performance using several imbalanced score metrics.
