Research And Application Of Support Vector Machine Based On Improved Ant Colony Algorithm

The support vector machine has an ideal data classification ability,but it affects the final classification result because of the difference of its optimal parameter.Therefore,this thesis puts forward the research and application of support vector machine based on improved ant colony algorithm.The classification accuracy and convergence speed are improved.Firstly,this thesis introduces the principle and concept of support vector machine.The effects of penalty factors and kernel parameters on the classification of support vector machines are analyzed.Then the principle of the basic ant colony algorithm is introduced.Aiming at the defects of ant colony algorithm,the corresponding improvement is put forward.Finally,it is applied to support vector machine parameter optimization.And compare with several other algorithms.The comparison shows that the improved algorithm has better classification accuracy and convergence speed.The innovation points of this thesis are as follows:1.Exchange rules and local search.The search rules of ant colony algorithm are based on the content of pheromones,and the more the path of pheromones,the more ants choose them.This thesis adopts double populations to improve.When the pheromone concentrations are quite different,the exchange rules are implemented to avoid excessive pheromone precocity and stagnation problems.In the exchange rule,when any of the two populations produces a local optimum,they can jump out so that the entire algorithm can achieve a global optimum value.In addition,this thesis improves the local search rules in two populations.And the new local search rules are defined according to the error model of SVM.The final search decide whether to make a global search or local search.2.Update of the pheromone formula.Only constant updating of pheromones will allow the algorithm to iterate and generate the global optimal solution.In this thesis,a new updating formula is designed.Let the updated pheromones guide the ants to find the optimal value.Combined with local search and exchange rules,the efficiency of algorithm search is improved.3.Combination of improved algorithm and support vector machine.The improved algorithm is used in support vector machine to realize parameter optimization and improve the accuracy and speed of classification.In this thesis,the experimental uses multiple data sets of UCI.Then it is applied to the application of personal credit rating.The experimental results show that the improved algorithm has been improved in performance and classification accuracy.