Patients Stratification in Imbalanced Datasets: A Roadmap

Learning in an imbalanced context is characterized by high disproportion ratios of data instances number belonging to each class of the dataset. Attributing the correct class for each instance is well studied using supervised learning techniques. However, the examination of effects of the same phenomenon in unsupervised learning environments lags behind. Some of the main issues hindering the application of unsupervised learning techniques (clustering techniques) in an imbalanced data setting are highlighted. It also presents a solution to deal with the showcased issues. This solution evades the noticed drawbacks by employing another set of clustering algorithms while including them in an aggregated learning framework. This set of algorithms would be assessed by measures tailored to the nature of these techniques and to the unique constraints that the imbalanced learning environment imposes. The suggested framework is intended to be applied to the patients stratification problem.