Research Of Analysis And Anomaly Detection Based On Mechanical Insurance Cost

With the constant improvement of the health insurance system, the covering number has been comprehensive. Due to a variety of information asymmetry, fraud behaviors especially insurance fraud has been of frequent occurrence. At the same time, the arrival of the rapid aging of population threatened the health insurance fund. Therefore, the establishment of a perfect and efficient anti-fraud mechanism is imminent.The main research of this paper is to identify anomalies and irregularities in Medicare cost data using data mining techniques. The main works are data preprocessing, finding abnormal clustering behavior, extraction of abnormal explanatory rules and the implementation of anomaly detection system.Firstly, in terms of feature analysis and extraction, the expenditure of one city’s health bureau was focused on. Respectively to do data filtering, data cleaning, and specifications, and carried out data warehouse management in accordance with the different application scenarios demand.Secondly, as to the discovery of abnormal clustering behaviors, the common types of diseases were selected as a representative for clustering analysis, then divided the possible abnormal samples by setting a threshold. And expert advice was combined with tag information to evaluate and optimize on the clustering results.Thirdly, in explanatory processing, using the rule extraction algorithm, excavated and extracted the reason in clustering exceptional class, then excavated in form of rules and store. At the same time got experts confirmation and support on the extraction rules.Finally, as to the implementation of anomaly detection system, this paper used a web development framework and developed an anomaly detection system that supports rules management, exceptions review, feedback and perspective analysis.Compared with the conventional anomaly detection based on supervised learning, unsupervised learning no longer over-rely on the core tag data. While adding the interpretative process and feedback mechanisms, effectively reduced the workload of healthcare auditors, more conducive to cycling operation of the system. Finally, the system provides a perspective analysis that can assist health care expert decision-making.