The Research On Large-scale Support Vector Machine And The Applications

With the development of informatization,massive data are produced rapidly in people's live.The value of big data is inexhaustible,waiting to be explored.In the bio-medical field,the data is growing in an explosive way.Researches show that the analysis to medical big data compares to an experienced doctor.Because human can't handle huge amounts of data,they need to research how to realize the intelligent data processing and mine the knowledge in data fully.Usually we use the knowledge of machine learning in data processing.Support vector machines(SVM)is a theory of machine learning for small data sets and plays an important role in the field of big data applications.In recent years,the text categorization algorithms and medical image classification algorithms based on support vector machine have been gradually applied.But usually the algorithms need a lot of repeated iteration and complicated calculation.With the increasing of the amount of data,the memory space increases greatly and sometimes can't satisfy the operation of the algorithm,which limits the use of SVM.The traditional solution is to change a large convex optimization problem into a series of small convex optimization problems.But the algorithms proposed based on this theory in dealing with large-scale data sets run slow.A new SVM method based on surrogate function,which would solve iteratively quadratic programming optimization problem by constructing surrogate function,is introduced in this paper.Introducing surrogate function can not only realize the parallel computation,but also guarantee the monotone decreasing of cost function and reduce the complexity of iteration.The update rule would be derived from three properties of the surrogate function and realized in MATLAB.In dealing with the convex quadratic programming problem,the proof of global convergence is presented because the global optimal solution must satisfy the KT condition.In the experimental stage,using the large-scale data sets from data generator and UCI database to test the SVM classification algorithm based on surrogate function,LIB SVM and SVMlight.By the three groups of experiments we finally draw the conclusion:when dealing with big data classification the SVM algorithm based on surrogate function shows an efficiently result.Finally,a group of medical large-scale text data after processing and transformation was chosen to test the three algorithms.By experiment contrast,the practicality and superiority of the S VM algorithm based on the surrogate function was further verified.