计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (21): 237-241.DOI: 10.3778/j.issn.1002-8331.1908-0165

• 工程与应用 • 上一篇    下一篇

基于TLSTM的医疗保险欺诈检测

曹鲁慧,秦丰林,闫中敏   

  1. 1.山东大学 信息化工作办公室,济南 250100
    2.山东大学 软件学院,济南 250100
  • 出版日期:2020-11-01 发布日期:2020-11-03

TLSTM-Based Medical Insurance Fraud Detection

CAO Luhui, QIN Fenglin, YAN Zhongmin   

  1. 1.Information Office, Shandong University, Jinan 250100, China
    2.School of Software, Shandong University, Jinan 250100, China
  • Online:2020-11-01 Published:2020-11-03

摘要:

医疗保险欺诈对医疗基金的正确使用造成了严重威胁。随着信息化的发展,越来越多的用户属性信息和行为信息被积累下来,使得通过分析用户行为序列进行欺诈识别成为了可能。但在医疗保险背景下,由于供需双方存在严重的信息不对称现象,欺诈者会努力模仿合法用户的行为,而且欺诈者的比例很小,传统的基于分类的欺诈识别算法不再适用。此外,患者的就医行为具有一定的偶发性,时间分布不均匀。针对样本不平衡和时间分布不均匀的挑战,提出基于TLSTM的医保欺诈识别框架,将用户的历史就医行为序列作为TLSTM模型的输入,预测患者再入院原因及诊疗方案,通过比较模型输出与用户当前就医行为的差异程度,来判断用户存在欺诈的可能性。实验表明,该算法在欺诈识别准确度上明显优于已有算法。

关键词: TLSTM算法, 欺诈识别, 行为模式

Abstract:

Medical insurance fraud is a serious threat to the proper use of medical fund. With the development of information technology, more and more user attribute information and behavior information that make it possible to detect fraud by analyzing user behavior sequences are accumulated. However, in the context of medical insurance, fraudsters will try to imitate the behavior of legitimate users because of the serious information asymmetry between the supply and demand sides, and the proportion of fraudsters is small. The traditional classification-based fraud identification algorithm is no longer applicable. In addition, the patient’s medical behavior has certain sporadic and uneven time distribution. Aiming at the challenges of sample imbalance and uneven time distribution, this paper proposes a TLSTM-based medical insurance fraud identification framework, which takes the user’s historical medical treatment sequence as the input of the TLSTM model, predicts the reasons for the patient’s readmission and the diagnosis and treatment plan, and compares the model output with the user. The degree of difference in current medical treatment behavior to determine the possibility of fraud. Experiments show that the proposed algorithm is superior to the existing algorithm in fraud recognition accuracy.

Key words: TLSTM algorithm, fraud identification, behavioral pattern