Classification Problems Research Based On Evolutionary Algorithms

Classification is a basic problem in machine learning field.It is significant to solve classification problems accurately and efficiently for scientific and engineering fields.As evolutionary algorithms(EAs)are applied in classification problems,the accuracy and speed of convergence for classification have been greatly improved.Compared with other traditional methods,EAs have excellent global optimization ability and robustness.Besides,EAs have the characteristics of strong expansibility and superior self-learning ability,so they can deal with some complicated problems adaptively.Fireworks algorithm(FWA)is an optimization algorithm which simulates fireworks exploding in the night sky.FWA has a good balance between exploration ability and exploitation ability.However,there are defects in FWA,such as good solutions in the swarm do not interact well in the process of optimization.Meanwhile,EAs are only used to improve the performance of classifiers either by optimizing the parameters or structure of the classifiers,or by pre-processing the inputs of the classifiers.In order to overcome the limitations of EAs in classification problem,improve optimization ability of FWA to deal with more complex problems,and expand the method of multi-objective classification,the primary research work of this paper is as follows:(1)First,three new optimization classification model have been proposed.In this model,classification problem is transformed into an optimization problem so that any EA can be applied to solve classification problems directly through the classification model.The experiments prove that all the optimization classification models can effectively predict the class label of the test data.(2)FWA is simple and its single search strategy cannot solve many kinds of complex practical problems.Thus,a self-adaptive fireworks algorithm(SaFWA)and an improved self-adaptive fireworks algorithm(ISaFWA)have been put forward.In SaFWA and ISaFWA,differential evolution(DE)search strategies are introduced into Gaussian sparks strategy to increase the interaction between other excellent solutions.The increase of DE strategies not only optimizes the diversity of the solutions,but also improves the efficiency of solving problems.Besides,a self-adaptive mechanism can dynamically choose the best search strategy for the corresponding problem during the search process,which can improve the universality and robustness of the algorithm.The experimental results show that SaFWA and ISaFWA have good performance and greatly enhance the ability of optimization classification.(3)Finally,in order to extend the domain of classification problems,this paper increases the strategy to solve multi-objective classification problem.The Non-dominated Sorting Genetic Algorithm(NSGA-?)was used to carry out experimental analysis in classification.The experimental results show that NSGA-? is able to find good Pareto curve and has good classification accuracy and robustness.