Research On Outlier Detection Algorithm And Its Application In Abnormal Detection Of Clinical Prescription Data In Electronic Medical Records

Data mining is a technique for mining valuable and understandable information that is regular,prominent,and difficult to discover from a large amount of complex data.Outlier detection is one of the important topics in the field of data mining.With the increasing amount and complexity of data generated by real life and the Internet,new challenges are proposed for outlier detection.This paper hopes to solve the problems in outlier detection by studying the basic theory and algorithm of outlier detection,so that the improved outlier detection is more suitable for the detection of abnormal data of clinical prescriptions of electronic medical records,and can effectively used the large amount of data accumulated in the hospital’s informatization construction to detect unreasonable prescriptions in a timely manner.The main research contents of this paper are as follows:(1)The Self Organizing Maps clustering algorithm is improved.In order to accelerate the convergence speed of the algorithm,and reduce the influence of the order of input data on the training results,this paper add the influence of the winning coefficient on the weight adjustment in the weight adjustment formula in the training process.Then define the reference function which reflecting the degree of network convergence,adjust the learning rate function according to reference function,and introduce the winning coefficient into the learning rate function,so that the algorithm can adaptively find the current most suitable learning rate,accelerate convergence speed while ensuring accuracy.(2)The traditional LOF algorithm is improved,and the "friend relationship" model is introduced.In order to solve the problem that the LOF algorithm misinterprets normal data points as outliers due to the problem of data distribution of data sets in some cases,a local outlier factor algorithm based on inverse neighbor density is proposed,and the new local outlier factor is defined.(3)The improved SOM clustering algorithm is applied to the initial stage of the outlier detection algorithm,and the data set is clustered and pruned,most of the normal data located in the cluster center is removed,and the data of the data set to be detected is reduced.Then,the LOF algorithm based on reverse neighbor density proposed in this paper is used to detect the detected data set,which reduces the algorithm time cost while ensuring the accuracy of the algorithm identification.(4)In this paper,the improved algorithm is applied to the actual electronic medical record clinical prescription data,and the clinical prescription data is compiled to obtain the clinical prescription data matrix divided by the department.After preprocess the data set,the traditional algorithm and the improved algorithm proposed in this paper are applied to the actual electronic medical record clinical prescription data.Analyzing and comparing the detection results and efficiency of the two algorithms,the result proves that the improved algorithm proposed in this paper is efficient.