Research On Classification Algorithm Of Medical Diagnostic Data Based On Kernel Method

Because of the property of special heterogeneity,massively,complexity and security,medical data can't be used properly in data analysis and data mine,and which has many incomplete and inconsistent redundant data in the process of data acquisition and processing.At present,how to mine the useful information from large medical database by applying the intelligent learning algorithms,and contribute to the disease diagnosing and medical researching,which is becoming a hot spot in the era of big data.In the medical data mining,especially those on classifying high-dimensional data,too much property variable would lead to large amount of calculation and increase the time consumption of medical data mining in some cases.And the noise which contained in the large amounts of medical data will reduce the accuracy of data classification,even affect the final decision and analysis for medical.Thus,it is of great significance for medical data mining to reduce the amount of calculation data processing and improve the accuracy of data classification.This thesis has made some research on KPCA method and SVM algorithm in terms of medical data's application based on the theory of kernel method,which showed that the SVM algorithm is suitable for the high-dimension pattern of data analysis,however,whose computation and time cost of algorithm is relatively high in the process of dealing with the complex pattern,high dimension,small sample and high noise of medical data.And for the dimension-reduction method based on KPCA method,it always has strong sensitivity on the data noise while reducing the dimension and integrating the information of data effectively.Therefore,according to the skills of weakening the reconstruction error of dimensionreduction with KPCA method while weakening the sensitivity of data noise,the thesis proposes an improved KPCA method(KEPCA).Compared with KPCA method on the high noise data denoising performance and the effect of dimension-reduction,it shows that KEPCA method has better denoising ability and better performance than KPCA method on data dimension-reduction.And then the dissertation proposes an improved multiple kernel SVM classification method based on the idea of the feature weighted SVM classification method.Experiments show that EWSVM method improves effectively the classification performance compared with other methods.The dissertation experimentalized on the 7 groups medical data which are representative in the sample size,dimension of properties and the amount of data noise according to the model of combined KPCA method with SVM method.The accuracy,sensitivity and specificity of the classificational model were taken as indicators to compare and analysis.And the experiments show that the proposed methods have better rationality and effectiveness on the classification accuracy when denoising,reducing the dimension of the classification dataand improving the classification accuracy,and which further proves that the model combined KPCA method with SVM of classification method has stronger ability of data classification.