| In recent years, with the rapid growth of information and data, the machine learning model has been widely promoted and applied. However,as the most crucial part of model training, the parameter optimization of the model has always been a difficult problem, and often takes a long project time. Classical optimization method spend a relatively large time complexity or space complexity, the effect of the final optimization depends on the function form of the problem. As a kind of stochastic search algorithm which imitates the behavior of biology groups, the group intelligent algorithm has been widely concerned and applied since it was put forward.In this research, the implementation steps and application scenarios of multi-population intelligent algorithm are introduced, including the origin,internal mechanism and improvement of PSO algorithm, the process of GA and immune algorithm and there characteristics are briefly introduced. In the face of the problem of multimodal function, the particle swarm algorithm with global topology leads to fast convergence and poor accuracy. Many researchers use local topological structure or dynamic topology to improve the diversity of particle swarm optimization algorithm,which are able to achieve good results. In this paper, a dynamic multi-group particle swarm optimization algorithm (DMGPSO) is proposed, which is based on the local topology PSO algorithm and the mechanism of dynamic grouping. The simulation results show that the proposed algorithm is better than the standard particle swarm algorithm (PSO) in solving multimodal function problems.Finally, the hybrid particle swarm optimization algorithm (GSA +DMGPSO) is used to optimize the model parameters of the breast cancer prediction classifier and liver disease classifier by combining DMGPSO and the grid search algorithm (GSA). These models use support vector machine with Gaussian kernel function, and the final models can achieve better classification accuracy. |
PDF Full Text Request