Optimization Of Nearest Neighbors Method In Dynamic Medical Expert System

With the rapid development of Internet technology,researchers pay attention to the related theory and application research of smart medicine which combines Data Mining technology and Expert System technology.Now,in the related research of smart medicine,the supporting data are usually static.While the medical knowledge and rules have the characteristics of dynamic change.So I make a dynamic research on the supporting data.Diagnostic or predictive models in smart medicine often require classification algorithms.So aiming at the classification algorithm of Data Mining,that is Nearest Neighbors method,I carried out a series of improvements,including the improvement of classification accuracy,the satisfaction of the samples quality demand,the convergence of time and the adaptation of dynamics.Meanwhile,in order to improve the classification accuracy,I also optimizes the similarity measurement algorithm which combine Euclidean Distance and Jaccard Distance and I call it as Euclidean-Jaccard Distance(E-JD).The improvement of Nearest Neighbors method has the following five stages.1)I proposed Sub-group Nearest Neighbors(SNN)method which is an improvement method of Nearest Neighbors.SNN method can eliminate problems of Nearest Neighbors method,that is the boundary ambiguity and the classification error caused by the uneven number of comparison samples.2)SNN method needs high-quality comparative sample set.Therefore,Dynamic Sub-group Nearest Neighbors(D-SNN)method is proposed.3)D-SNN method has the problem of increasing time consumption as the number of samples increases,so Time-convergentDynamic Sub-group Nearest Neighbors(TD-SNN)method is proposed.4)TD-SNN method sacrifices the accuracy.Therefore,Fault-tolerant Dynamic Sub-group Nearest Neighbors(FD-SNN)method is proposed to achieve time convergence and ensure the classification accuracy by adjusting fault-tolerant parameters.5)In order to adapt to the dynamics better,I further improve the FD-SNN method by adding the automatic assignment method,that is propose Fault-tolerant Dynamic Sub-group Nearest Neighbor(AFD-SNN)method.At the same time,I improve the similarity measurement algorithm,mainly including the improvement of distance algorithm,that is,the combination of Euclidean Distance and Jaccard Distance.Experiments show that SNN method is more effective and stabler than Nearest Neighbors method.FD-SNN method achieves time convergence and ensures classification accuracy.The classification results of AFD-SNN method is not optimal,and its automatic assignment method needs further study.The results of E-JD is better than the single distance's.And the E-JD formula 2 is better than others.The classification effect of the two data transformation operations is similar.