Research On SVM Modeling In Medical Data Classification

In recent years, with the development of medical technology and therising popularity of medical information technology, especially withhospital widely naturalized in a variety of electronic medical equipment,make more and more medical data, and presents the tendency of rapidgrowth, marked the medical data has entered the era of big data. How tomake use of information technology from these data found useful orunknown information, provide doctors with auxiliary medical diagnosishas become an important direction of the research at present.SVM as a good data classification method, based on characteristicvalue can be used in pathology classification, however many artificialfactors exist in the process of its modeling interventions, such as thekernel function selection, constant parameters, etc. Will take breast cancerdata as the application background, this paper is the modeling process ofSVM classification model and kernel function selection, parameter forstudy, this paper mainly includes the following contents:(1) Analysis of the current commonly used methods of the SVM andBP neural network and decision tree classification principle, this paper introduces the particle swarm optimization (PSO) algorithm forparameters optimization of the mechanism, and related theory of nuclearparameters.(2) Studied the modeling process of the SVM model, established theoptimization model of classifying breast tumor. In view of the normalizeddata set problems are discussed in this paper, analyzes the datanormalization on the result of classification. The choice of kernelfunction is discussed problem, verify the classification of the radial basiskernel function has obvious effect. For the choice of model parameters Cand g, adopted a gradual intensification of quadratic optimization method,for the first time from the parameters of rough, within the scope of thefine for the first time found relatively optimized parameters. Actual testresults show that the model has good classification accuracy.(3) The particle swarm optimization (PSO) algorithm was used tooptimize the SVM model parameters are studied. The traditional particleswarm optimization (PSO) punishment for SVM parameters C and goptimization method, in view of the traditional algorithm learning factorby empirical value, this paper introduces a dynamic modification factorof the improved algorithm study. Through experiments on theclassification effect of these two methods were compared, the resultsshow that the modified learning factor of particle swarm optimization(PSO) algorithm in dynamic optimization of the classification model has the advantages of shorter operation time.