Volume 29, Issue 4, 2020
DOI: 10.24205/03276716.2020.855
Research on Active Discovery Model of Medical Insurance Fraud
Abstract
Medical insurance is a major issue related to the national economy, people's livelihood and national development. The problem of medical insurance fraud seriously threatens the safety of medical insurance funds and hinders the effective implementation of medical insurance policies. Therefore, the active detection of medical insurance fraud is of great importance to the development, improvement and social stability of medical insurance. significance. This paper presents a method of identifying fraudulent behaviors based on BP neural network. For data processing, we chose Excel and Access to summarize and normalize the patient data in Table 1 and the cost schedule in Table 2 based on the patient ID, and eliminate invalid data including incomplete records and format errors. In the process, we discovered that all the consumption records were only for the purchase of medicines, and in this month's consumption records, only a very small number of patients had the behavior of transferring to the department, and some patients were paid at their own expense, and there was no suspected medical insurance fraud, and some patients shared by multiple people. The phenomenon of the medical insurance card directly determines that it is a medical insurance fraud. The consumption records of these patients provide sample support for us to train the BP neural network. For this question, we first used Excel and Access to filter out information useful for fraud identification from a large amount of data, including the patient's age, gender, department, total consumption of the current month, and consumption frequency of the current month, etc. You, and consider There are differences in the consumption situation of different departments, so we calculated the average consumption of each department, and made the relative difference between the consumption of each patient in the current month and the average consumption of the corresponding department. With these fraud factors and the consumption records of self-paid patients and patients who share medical insurance cards, we established a Logistic binary regression model to evaluate the impact of each fraud factor on the probability of fraud and eliminate fraud factors that are invalid for the possibility of fraud. , The fraud factors that have significant influence on the possibility of fraud are retained as the input vector to train the BP nerve, and the trained network is used to identify the fraud of medical insurance patients. In the end, we believe that the patient whose output is 1 is suspected of major medical insurance fraud.
Keywords
Medical insurance fraud, Logistic binary regression, BP neural network data