Abnormal Behavior Detection Algorithm And System For Medicare Date

With the development of health insurance and widespread use of health insurance cards, noncompliant behavior of Medicare card use is becoming more common, such as health care fraud, Medicare card cash, excessive treatment, etc. Traditionally, the main method is to use a rule set. The accuracy of this method is low, as the rules are developed artificial. With the rapid development of machine learning and pattern recognition, anomaly detection technology has made great breakthrough. Using outlier detection techniques to identify abnormal behavior from Medicare data set, can improve the accuracy and recall rate of abnormal behavior detection, and has important significance and application value.In this paper, we start the study of abnormal behavioral techniques for Medicare data, the main contents are:① First, we introduce anomaly detection technology, and summarize several reasons for outliers. Compared with the traditional rule-based anomaly detection method, we elaborate on anomaly detections based on classification, clustering, distance, and statistics, including their advantages and disadvantages and applicability.② As the basis of abnormal detection algorithm based on clustering, we design and implement an on-line outlier detection algorithm:Incremental Cluster-Based LOF. The algorithm first cluster the data set, and then calculate the degree of abnormal values for each data object. At last, we select some objects which get the highest score values as outliers. The basic idea of the algorithm is based on the degree of deviation from a data object and the adjacent points, to determine whether the data is abnormal. This model can detect the global and local outliers.③ We propose an online anomaly detection algorithm:BFS (Breadth-First-Search) SmartSifter. The algorithm takes the data in different cell according to the discrete attributes of data, and’then constructs a Gaussian mixture model for each cell. The basic idea of the algorithm is based on the degree of change in the data object model, to determine whether the data object is abnormal.④ We implement online abnormal behavior detection system for Medicare data, and introduce the system structure and some important module, and provide a friendly interactive interface.