AI Sleuth for Billing in Nursing Care
The billing documents are also an interplay of service records, tour schedules, duty rosters and other documents – these must be combined during the audit in order to uncover fraud. »A conspicuous feature can be, for example, that many of the nurse's services were billed at the same time in the service record, but the duty roster only lists a short assignment. We have to find such peculiarities in an automated way,« says Dr. Elisabeth Leoff, deputy head of the »Financial Mathematics« department at Fraunhofer ITWM.
»Our joint research project is intended to shorten the time-consuming and labor-intensive evaluation of evidence by shifting the manual to an automated collection and evaluation of care service documents,« explains Steffen Leitte, public prosecutor at the Dresden General Prosecutor's Office. »This requires AI expertise and mathematics.«
Machine Learning (ML) methods are used in the research project, especially in the digital capture of the various document types. A combination of image processing methods and modern Deep Learning methods is used. Various algorithms learn from a mixture of artificial and anonymized real data first to recognize crucial information and then to detect anomalies. In order to train these AI algorithms, a database is filled by the ITWM team and the Leipzig police department. This means that several thousand documents must have been created by humans and marked with properties in order to make the algorithm intelligent in the first place. The algorithms are programmed and repeatedly tested and improved with data from real investigative procedures. The analysis of the documents then forms the basis for the evaluation and detection of conspicuous features.