Research On Incremental Learning Algorithm Of Support Vector Machine

Support Vector Machine(SVM)is a completely new machine learning method proposed by Vapnik in the 1990 s for classification and regression.Based on statistical learning theory and structural risk minimization principle,it shows many unique advantages in solving the problem of small sample,nonlinear and high-dimensional pattern learning.It has been widely used in various fields.Twin Vector Machine(TWSVM)is a machine learning method based on SVM.It is formally similar to SVM but its computation time is one quarter of SVM.With the development of technology and information age,there will be a huge amount of data that needs us to handle.It is very difficult to obtain and process all the information at one time.The emergence of incremental learning technology solves such problems.It can continue to accumulate and increase with the data and improve learning accuracy.Incremental learning techniques perfectly handle historical results,preserving and utilizing it,and reducing training time for new samples through the use of historical results.Therefore,it is necessary to study support vector machine incremental learning algorithm.The main work includes:(1)Research machine learning problems and statistical learning theory into,analysis of the advantages and disadvantages of support vector machines.(2)The incremental learning algorithm of support vector machine is introduced and the concept of pre-selection of boundary vectors is introduced.An incremental learning algorithm of support vector machine based on K-nearest neighbor center density is proposed for dichotomous problems.The performance of the algorithm is tested on the UCI dataset.Experimental results show that compared with the standard SVM training algorithm,the training speed has been significantly improved in the case of small impact on accuracy.(3)Aiming at the problem of multi-class classification,a new classification model is proposed based on twin SVM,and a new incremental learning algorithm Experiments show that the new incremental learning algorithm improves the training efficiency under the condition of little impact on the training accuracy.