Developing an anomaly detection framework for Medicare claims

Detection rates of non-compliant activity in Australian provider medical claims are below international benchmarks, and new methods are required. Since the Department of Health requires interpretability and incorporation of expert feedback as key components of its decision support systems many existing fraud detection techniques are unsuitable. As part of an Industry PhD project we have developed several new anomaly detection techniques which have been implemented in a prototype software system for the Australian Government Department of Health. We discuss the goals of the project and outline our approach to rapid prototyping and implementing processes to achieve expert-validated improvements in detection rates.