A Class Of Online Algorithms For Support Vector Machine And Their Application

Support vector machine(SVM)is a classic technology for classification,which has been widely used in various fields due to its effectiveness and feasibility.However,as the scale of data continues to grow,the need for real-time data processing has become more and more urgent.Traditional SVM algorithms have encountered bottlenecks in not only computational efficiency but also computational cost.In order to solve this problem,we first use the penalty function method to convert the original SVM optimization problem into an unconstrained one,and then apply the stochastic proximal gradient method to solve the problem,which results in an online algorithm for SVM(denoted by SPG-SVM).According to whether the amount of data used in the iteration is single or batch,we divide SPG-SVM into SPG-SVM(single point)and SPG-SVM(batch).Then based on the idea of Passive-Aggressive(PA)algorithm,combined with the penalty function method,we propose a stochastic proximal gradient method to solve the same problem,the online algorithm is called SPG-PA.Finally,we compare the performance of the traditional SVM,SPG-SVM(single point),SPG-SVM(batch),PA and SPG-PA algorithms on the data set of the outbound purchase of marketing products of an operator in a certain province.The results show that the online algorithms have obvious advantages in the running time of model training compared with the traditional offline algorithm.Meanwhile,SPG-SVM(single point)and SPG-SVM(batch)shows excellent property in forecast of the purchase of marketing products.