关键词: 过程挖掘, 临床诊疗过程, 异常发现, 合规性检查, 对齐成本

Abstract: To adequately use information in the health information system data for abstracting clinical process reasonably,and to detect explainable and exact outliers,based on considering the semantic information,order information and frequency information in the data,an improved method to detect outliers was proposed,which assumed that most healthcare behaviors conducted by doctors were normal,only a few behaviors deviated from that were outliers.Latent Dirichlet Allocation (LDA) model was used to cluster billing items,so that the topic distribution of each day of patients was obtained;based on the topic distribution obtained above,K-means++ was used to cluster all the days;based on the clusters obtained above,Inductive Miner-infrequent (IMi) was used to discover a sound 80% model.Alignment-based conformance checking technology was used to align the log and model,so that the positions and types of outliers occur was got,also the extent to which the log deviated from the 80% model.The experiment results showed that the proposed method could produce more explainable and more exact outliers,and could support the investigation of health care insurance.

Key words: process mining, clinical process, outlier detection, conformance checking, alignment cost