Research On SVM Classification Algorithm Merged With Fisher Discriminant Analysis

SVM is a supervised learning model,which can be used for data classification,regression analysis and outlier monitoring.It is based on the principle of VC-dimensional structural risk minimization,and is widely used in the field of model identification and artificial intelligence.As a complete supervised binary classification model for machine learning,an optimization algorithm for convex quadratic programming is constructed by maximizing the interval in feature space.To solve the nonlinear separable problem,SVM transforms the nonlinear problem into the indivisible problem in Hilbert space via kernel mapping.In this paper,we focus on the binary classification Support Vector Machine,merged with Fisher Discriminate Analysis,making for the limitations of SVM,in order to improve the accuracy of the classification of sample data as a result.Firstly,we use FDA based on weighted SVs to optimize the target equation of SVM,which improves the accuracy of SVM classifier for data with different dimensions.Secondly,in the face of unbalanced SVM classification problem,we adopt an over-sampling technique that combines FDA's rules,preprocessing the unbalanced binary classification data with an iterative algorithm,clustering algorithm and GMM algorithm are used in the process.Finally,the new over-sampling technique is used in clustered SVM to improve the accuracy of classification.The above two improved algorithms are tested in manual dataset and real dataset,which get apparent good results compared with historical approaches.