Research On Fraud Detection Of Medical Insurance

Medical insurance is a prerequisite for payment of insurance under the convention. Moreover, it insured with medical expenses of diagnosis and treatment during the period of insurance service. The occurrence of medical behavior under the insurance contract. It can protect patient medical treatment from the economic losses in medical. Medical insurance is a national service for people’s lives, an important measure to protect people’s health, which involves in the vital interests of each insured. On the contrary, driven by interests, a long-term problem of medical insurance fraud exists in insurance, moreover the new fraud is endless. The loss of insurance is also a damage to the interests of the insured. Hence it hinders the construction of medical insurance system.Fraud and anti-fraud become a chess phenomenon. How to improve anti-fraud is a long-term study. This thesis begin with data analysis, digging out valuable information of medical insurance data using big-data technology, finding the potential rule, to achieve the purpose of detecting medical insurance fraud. Medical insurance data reflect more hidden behavior of the participants and the organization. Therefore, it is necessary to find out the regularity of related and conflicting behaviors and relationships in order to assist superintendent for a more efficient use and management.This thesis study on Medical insurance fraud detection. It burrows deep into the insured and medical behavior information, aimed of finding consistent with "normal" and "abnormal" Medicare card usage behavior, and the discovery of prescription, treatment and whether the link with conventional pharmacological treatment methods two scenarios. The main contents include as followed.1) It frequently dug insurance card records, and find frequent item sets and non-frequent item sets according to the emergence of more and fewer records mining semantic conflicts, then creates a large event graph. Through events graph clustering, it can obtained "absolute conflict "and" suspected conflict "pattern. On "suspected conflict" cluster may be further determined by the study are "absolute conflict" areas of behavior, and gradually improve the behavior patterns of semantic mining. Experiments show that the method based on actual data records has a high precision which helps with anti-fault in insurance. Conflict detection mode can effectively suspected of "abnormal" behavior insurance card use.2) Using LDA to model prescription content, drug name, disease modeling. The kind of text records, prescription content, text information drug names are mapped to the condition relating to the type of space by analyzing the (prescription drugs) Binary implicit condition for category factor for drug prescription process modeling organization. LDA obtain a probability of each drug name appear in the training set prescription, comparing with reality prescription, and the calculated pharmacopoeia library, introducing to the penalty parameter Г through the results by comparing the rigorous scientific description of pharmacological drugs, in order to get more accurate results. Experiments show that adding pharmacological contrast gets higher precision and more conducive to the analysis of insurance fraud. This algorithm can discover whether prescription meet the requirements and the presence of illegal prescription.