Research On Brainstorming Optimization Algorithms For Solving Classification Problems

Classification is the most basic and representative problem in the field of data mining and machine learning.Accurate and efficient classification is the basis of many scientific research and application engineering.Evolutionary computation(EC)technology has been successfully applied to solve many problems,such as classification problem,feature selection problem and so on,because of their good global optimization ability.Evolutionary classification model is one of the methods to solve classification problems.In recent years,brain storm optimization(BSO)has been used to implement the evolutionary classification model,and achieved the desired results,which means that it is feasible to solve the classification problem directly by using EC techniques.However,the existing evolutionary classification models are still insufficient.For example,the structure of evolutionary classification model is closely related to the dimension of dataset,but there are a lot of redundant,irrelevant or noisy features in the actual dataset,which may lead to complex calculation and low learning efficiency,thus affecting the classification performance of the model.In addition,in the optimization process,BSO algorithm cannot keep the information of the dominant solution,which may also lead to its classification performance is limited.In order to optimize the performance of evolutionary classification model from different angles and solve the classification problem better,this paper mainly does the following research work:(1)The complexity of evolutionary classification model structure depends on the dimension of data set.In order to optimize the structure of evolutionary classification model and improve the performance of classification model,this paper introduces the concept of feature selection and proposes two evolutionary classification models based on feature selection: evolutionary classification model based on structure and evolutionary classification model based on structure and its weight parameters.1)The structure based evolutionary classification model takes the highest classification accuracy as the objective function,and uses the global optimization ability of BSO algorithm to continuously search the optimal feature subset(model structure)with higher classification accuracy and smaller solution,so as to ensure the classification performance and reduce the complexity of the model structure.2)Different feature subsets have their corresponding optimal weight parameters,but in the process of searching the optimal feature subset,the corresponding optimal weight parameters are not searched,which easily leads to the loss of the better feature subset.On this basis,an evolutionary classification model based on structure and its weight parameters is proposed.On the one hand,BSO algorithm is used to search the optimal model structure;On the other hand,the BSO algorithm is used to search the optimal weight parameters under different structures to avoid discarding the better feature subset due to the mismatch of weight parameters.In this paper,the experimental results show the effectiveness of the two methods which introduce feature selection to reduce the model structure and improve the performance of model classification.(2)Aiming at the limitation of BSO algorithm,the search strategy of BSO algorithm is improved,and an improved brain storm optimization(IBSO)algorithm is proposed.In IBSO algorithm,by retaining the brainstorming process of BSO algorithm,the global optimal,local optimal and nearest neighbor search strategies are embedded into BSO algorithm,which can not only better retain the information of the dominant solution,but also increase the information exchange between the dominant solutions,and improve the local development and global exploration ability of the algorithm.In addition,in the process of searching the optimal feature subset for BSO,we use four transfer functions to constrain the range of feature subset to be in the binary value range.The performance of BSO algorithm is improved in the experiments.