Research On Instance Selection Method For K Neighborhood Classification

Nearest neighbor classification is one of the important research contents of machine learning.Among them,the K nearest neighbor classification algorithm is a very typical non-parametric lazy learning method,because of its simple and effective characteristics,it is widely used in various fields.However,its disadvantages have gradually emerged,including the large amount of redundant and noisy data in the data set,which will seriously affect the classification accuracy;when faced large data sets or data sets with high dimensions,the calculation cost is huge.This requires data pre-processing to improve the overall quality of the data samples,where data reduction is one of the effective means of data pre-processing.Instance selection as a commonly used data reduction method,it can effectively alleviate the above drawbacks.Existing instance selection algorithms for K-nearest neighbor classification have achieved many results,but there are still some shortcomings such as accidentally deleting non-redundant and non-noisy samples in the data set,and inefficient algorithms in the face of large-scale data sets.In order to improve the classification efficiency of the K-nearest neighbor classifier,this paper proposes two corresponding intelligent optimization algorithms through the research and analysis of many example selection algorithms and the combination of evolutionary calculation.The main research work is as follows:1.Summarize some related definitions and problems involved in the selection of examples.Then,different classification methods of instance selection algorithm are given,the relationship between instance selection problem and neighbor classification is explained,and the traditional instance selection algorithm for neighbor classification is briefly introduced.Then it introduces its process of solving the instance selection problem from the perspective of evolutionary algorithm and introduces the corresponding evolutionary instance algorithm model in detail.Finally,other related technologies that can improve the effect of the evolutionary instance selection algorithm are briefly explained.2.Aiming at the problem of redundant and noisy data affecting the classification performance in the instance selection process for K-nearest neighbor classification,in this paper,this paper proposes a cooperative selection algorithm(NNC-CoCo)for K-nearest neighbor classification.In the evolutionary instance selection process of the algorithm,a multi-point crossover strategy is used to further improve the accuracy of instance selection.At the same time,a fast mutation strategy is used for instance weighting and feature weighting,and collaboratively cooperates with instance selection to remove noise and redundant instances.The best training subset is obtained,and finally the performance of the K-nearest neighbor classifier is improved.Experimental results show that this method has advantages in classification accuracy and efficiency compared with some current evolution instance selection algorithms.3.In some datasets with a large number of instances,the time required for the evolutionary instance selection algorithm increases exponentially and the algorithm is inefficient.This paper proposes a hierarchical evolution instance selection algorithm(EIS)for K neighbor classification.Firstly,divide the data set using the idea of stratified random sampling,followed by the evolution of two independent populations,and the strategy based on local search mutation and the strategy based on elite individual replacement are used to improve the standard genetic algorithm and find a suitable training subset for correct classification.This method has advantages in the classification efficiency and storage rate of some current classic instance selection algorithms when facing some datasets with a large number of instances.