医保欺诈行为的主动发现——基于熵权法引入指标权重的聚类分析算法
Proactive Detection of Health Insurance Fraud—Clustering Analysis Algorithm Based on Entropy Weighting Method Introducing Indicator Weights

本文针对一些现有的识别方法存在的问题进行改进，应用“基于熵权法引进指标权重的聚类分析算法”进行医保欺诈行为的识别。在完全舍弃主观赋权的情况下，精确地识别出发生欺诈行为的个案。首先对无意义的数据进行降维，并结合医疗保险欺诈的现实案例，综合筛选出五个指标，而后引入信息熵的概念，并基于熵权法确定指标权重；为了避免各个指标的权重给定中存在的主观性，本文通过对信息熵的刻画来体现某一个指标所拥有的信息量期望，最后将得到的权重应用于“改进的欧式距离”，通过对不同指标的“距离”进行赋权，得到一种全新的“距离”用于聚类分析。并按照账单号合并多条拿药记录，以账单号为索引，通过层次聚类分析算法构建聚类树。本文认定：医保欺诈行为是完全地呈孤立点分布的。通过改变聚类数，得到不同聚类数下的孤立点个数，最终结合相关案例，选定聚类数为4，由此求得并给出疑似发生医保欺诈的账单记录43个。
This paper addresses the problems of some existing identification methods and applies a “clustering analysis algorithm based on entropic weighting to introduce indicator weights”. In order to avoid subjectivity in the weighting of each indicator, this paper uses the information entropy to characterise the expected information content of a particular indicator. In order to avoid the subjectivity in the weighting of each indicator, this paper reflects the information expectation of a certain indicator through the portrayal of information entropy, and finally applies the obtained weights to the “improved Euclidean distance”, and obtains a new “distance” by assigning weights to the “distances” of different indicators for clustering analysis. The paper also merges multiple medication taking records by billing number and uses the billing number as the index to construct a clustering tree by a hierarchical cluster analysis algorithm. This paper concludes that: health insurance fraud is completely distributed in isolated points. By varying the number of clusters, the number of isolated points under different clusters was obtained, and finally the number of clusters was selected to be 4 in conjunction with relevant cases, which resulted in 43 billing records suspected to have been fraudulent.