Parameter Optimization of SVM Based on Improved Levy Flying Particle Swarm Optimization Algorithm

Abstract

The invention relates to a problem of combining a data mining algorithm and a random search algorithm, and proposes an SVM parameter optimization algorithm based on an improved Levy flying particle swarm algorithm. With the advent of the era of big data, how to improve the performance of data mining algorithms has become a research hotspot. When the parameters of SVM are optimized by the traditional PSO algorithm, it is easy to fall into the local extremum, and the later oscillation is serious. Levy flight combines short-range searching with occasional long-range walking, which can fundamentally overcome the problem that PSO algorithm is easy to fall into local extremum. Considering the serious oscillation in the later period of the algorithm, the improved particle position update formulaand the introduction of momentum term can reduce the oscillation of the algorithm. At the same time, considering that the value of the inertia weight trades off the local search and the global search,the adaptive inertia weight based on the distance between particles of the invention can improve the algorithm convergence speed, thereby improving the PSO algorithm searching ability, and further finding the optimal SVM classification model.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to parameter optimization of a support vector machine and an improved Levi's flight particle swarm algorithm. Background technique [0002] With the advent of the big data era, higher requirements are put forward for data analysis tools, and various data mining algorithms for big data have become research hotspots, such as support vector machines (Support Vector Machines, SVM), artificial neural networks (Artificial neural networks networks, ANN), decision tree (Decision tree, DT), etc. Support vector machine is one of the most widely used classification algorithms. The core idea of ​​this method is to find the optimal hyperplane in the feature space to separate the two types of samples without error, and the classification interval is the largest. The traditional support vector machine has a strong dependence on internal parameters, and the selection of the penalty parameter C...

AI Technical Summary

Problems solved by technology

[0003] In the traditional particle swarm optimization algorithm, as the number of iterations increases, the particles will appear "agglomeration", which will cause the algorithm to easily fall into local extremum. Considering the characteristics of Levi's flight can break this "agglomeration" phenomenon, the Levi's flight Introduced into the particle swarm algorithm, but the traditional Levy flight particle swarm algorithm simply combines the two, although it fundamentally overcomes the problem that the PSO algorithm is easy to fall into local extremum, but increases the complexity of the algorithm, the present invention Improvement on the basis of Levi's flight particle swarm algorithm to improve the search ability of the algorithm

[0004] Considering the problem of severe oscillation in the late stage of the algorithm, the present invention designs a new following particle

Images