Novelty detection using one-class Parzen density estimator. An application to surveillance of nosocomial infections

Abstract

Nosocomial infections (NIs) - those acquired in health care settings - represent one of the major causes of increased mortality in hospitalized patients. As they are a real problem for both patients and health authorities, the development of an effective surveillance system to monitor and detect them is of paramount importance. This paper presents a retrospective analysis of a prevalence survey of NIs done in the Geneva University Hospital. The objective is to identify patients with one or more NIs based on clinical and other data collected during the survey. In this classification task, the main difficulty lies in the significant imbalance between positive and negative cases. To overcome this problem, we investigate one-class Parzen density estimator which can be trained to differentiate two classes taking examples from a single class. The results obtained are encouraging: whereas standard 2-class SVMs scored a baseline sensitivity of 50.6% on this problem, the one-class approach increased sensitivity to as much as 88.6%. These results suggest that one-class Parzen density estimator can provide an effective and efficient way of overcoming data imbalance in classification problems.