Anomaly detection in the Brazilian public health system

Andrea Iabrudi Tavares

DECOM - UFMG

Anomaly detection is a common problem common in many applications areas such as fraud detection in financial transactions as the patterns on credit card purchases by customers. Other applications include the detection of anomalous behaviors of mobile users, and anomalies in user data in health plans. Data mining methods have been developed to deal with the aspects common to all theseapplications, as well as the specific aspects of each. Among the common axes that run through all the techniques, the main ones are: visualization as a tool for exploring hypotheses, multivariate analysis, and robustness as the way to ensure that the presence of abnormalities do not mask the overall statistics and prevent their detection.

In this presentation, we present the ongoing experience of a group from the Departamento de Ciencia da Computacao da UFMG to develop an anomaly detection system, called INFOSAS, to the Brazilian Health Ministry. In paryicular, we are interested in detecting frauds. We discuss some of the mahor fraud schemes and the techniques used to address them. Some of the generic conclusions we have reached at this point is the need to use simple and understandable models, visual tools, and a flexible environment that gives data management its pivoltal value. We also emphasize the need to build interactively this system with human intervention and heavy help from public health experts.