Detecting Healthcare Fraud Using Hybrid Machine Learning for Document Digitization

Fraud and abuse are significant and growing problems for healthcare systems worldwide. In Germany, nursing services perform a variety of billable activities, which are invoiced using various paper documents. Some of these are standardized but not firmly formatted; most of them combine printed and handwritten texts. In the event of anomalies or fraud, all this information must be combined. In many cases this is still done manually, which is immensely time-consuming. Our aim is to support German law enforcement agencies detecting and persecuting these cases. We developed a semi-automatic method for document digitization that combines various state-of-the-art machine learning techniques to not only give reliable results but enable user interference and traceability for nonexperts. We evaluate our solution with the help of legal authorities as well as investigators and apply it to data from real criminal investigations.