The Health Care Fraud Detection Using the Pharmacopoeia Spectrum Tree and Neural Network Analytic Contribution Hierarchy Process

摘要

Recent years, data mining has used frequently for medical fraud detection and neural network has its special merit. This paper proposed an improved neural network algorithm to detect the medical insurance fraud. Our method combined MPL neural network with neural network analytic contribution hierarchy process and make corresponding improvement according to the characteristics of the medical insurance fraud: (1) build a pharmacopoeia spectrum tree, using neural network analytic contribution hierarchy process to cluster the medical items to obtain reasonable fraud detection factor, (2) for each case, use the method of hierarchy contribution rate and multidimensional space distance to calculate the contribution rate of fraud the fraud detection factor have in the medical items, so as to find out the most possible fraud medical items. Our experiment results show that the accuracy of the improved neural network in health care fraud detection reached 86%, which is better than other unsupervised data mining methods. What is more, our method can calculate each fraud detection factor's contribution rate of fraud, which is the ability that other data mining methods do not have.