Research On Multiple Birth Support Vector Machines Models And Algorithms Based On Structural Information

Twin support vector machines have a wide range of applications in the field of machine learning,and their main goal is to solve binary classification problems.But the multi-classification problem is the most common in real life.Therefore,researchers have improved the twin support vector machines and proposed many different classifications of multi-class twin support vector machines to solve the multiclassification problem.As a new type of improved method,multiple support vector machines are limited in the size of each quadratic programming problem only by the data size of the corresponding class.Compared with other multi-class twin support vector machines,it has low computational complexity and training.Speed and other advantages.Therefore,it is favored by researchers.However,the multiple support vector machine ignores the structural information of the sample data when building the model,and this information may contain some important prior knowledge.At the same time,multiple support vector machines often rely on empirical values or grid search methods for parameter selection,which makes the algorithm easily fall into a local optimal solution.In this paper,the structured information of sample data is used to improve the multiple support vector machine.At the same time,in the parameter selection of the multiple support vector machine,the group intelligent optimization algorithm is used to optimize it to improve the classification accuracy of the algorithm.The main research contents are as follows:First,this paper studies the structured least square multiple support vector machine based on energy.The structured least squares twin support vector machine is improved,and an energy-based structured least squares twin support vector machine is proposed.It transforms the inequality constraints in the structured least squares twin support vector machine quadratic programming problem into energy-based equality constraints by introducing an energy factor for each hyperplane.The algorithm can reduce noise while reducing time complexity.The effect of points on the algorithm.Based on this,the energy-based structured least squares twin support vector machine is extended to the multi-classification problem,and an energy-based structured least squares multiple support vector machine is proposed.In order to verify the effectiveness of the algorithm,we tested it on the UCI dataset and used Friedman test and ROC analysis for statistical analysis.Experimental results show that the proposed algorithm has good classification performance.Next,this paper studies a structured multiple support vector machine based on a dynamic quantum particle swarm optimization algorithm.Based on the classical quantum particle swarm optimization algorithm,the individual particle search ability factor is defined,and it is used as the feedback information to dynamically adjust the CE coefficient.A dynamic quantum particle swarm optimization algorithm is proposed.Subsequently,this paper uses this algorithm in an energy-based structured least squares multiple support vector machine to optimize its parameters.The algorithm uses the excellent global search ability of the dynamic quantum particle swarm optimization algorithm,which not only has a fast search speed,but also prevents the algorithm from prematurely falling into a local optimal solution.It has good global convergence and avoids the blindness of parameter selection.The classification performance of energybased structured least squares multiple support vector machines is further improved.Finally,the effectiveness of the dynamic quantum particle swarm optimization algorithm and the structured least squares multiple support vector machine based on the dynamic quantum particle swarm optimization algorithm is tested experimentally.The experimental results show that the convergence effect of the dynamic quantum particle swarm optimization algorithm proposed in this paper is greatly improved compared with the classical quantum particle swarm optimization algorithm.Moreover,experiments on the UCI dataset show that the dynamic quantum particle swarm optimization algorithm can find more suitable parameters for the energy-based structured least squares multiple support vector machine,which improves the classification accuracy of the algorithm.This paper has 15 figures,17 tables and 80 references.