Computational intelligence contributions to readmisision risk prediction in Healthcare systems

Abstract

The Thesis tackles the problem of readmission risk prediction in healthcare systems from a machine learning and computational intelligence point of view. Readmission has been recognized as an indicator of healthcare quality with primary economic importance. We examine two specific instances of the problem, the emergency department (ED) admission and heart failure (HF) patient care using anonymized datasets from three institutions to carry real-life computational experiments validating the proposed approaches. The main difficulties posed by this kind of datasets is their high class imbalance ratio, and the lack of informative value of the recorded variables. This thesis reports the results of innovative class balancing approaches and new classification architectures.