Kernel Extreme Learning Machine Based On Improved Flexible Polyhedron Algorithm

The kernel extreme learning machine is developed by combining the kernel function with the extreme learning machine.It not only keeps the advantages of high accuracy and fast operation speed of traditional ELM classification,but also improves generalization ability and stability.However,if KELM chooses parameters that are not suitable,it will directly affect the classification accuracy of the model.To solve the problems of KELM algorithm,this paper proposes a method to self-adaptived optimize the parameters of the kernel extreme learning machine based on the improved flexible polyhedron algorithm.(1)To solve the problem of parameter optimization of KELM model,a flexible polyhedron is used to optimize the kernel parameter and penalty parameter.Firstly,the initial flexible polyhedron is provided by the grid-search optimal range,and then the weight value is added to the deformation search parameters to distinguish the influence of kernel parameter and penalty parameter on the classification performance of the kernel extreme learning machine.Finally,KELM model is constructed by the optimal parameters obtained by the iterative search of the flexible polyhedron.The feasibility of the algorithm is verified on the UCI public database and the artificial datasets.(2)In reality,a lot of datasets have the characteristics of unbalanced labels.In order to improve the performance of KELM model in unbalanced classification problems,this paper proposes a class-specific cost regulation kernel extreme learning machine based on improved algorithm of flexible polyhedron.The model of class-specific cost regulation kernel extreme learning machine contains minority class penalty parameter and majority class penalty parameter.Furthermore,the improved flexible polyhedron algorithm is used to optimize the key parameters of CCR-KELM.The feasibility of the algorithm was verified on KEEL unbalanced binary datasets.Finally,the algorithm is used to identify abnormal heart sound signals in the application of heart sound classification.