Research On Classification Learning Based On Rough Sets

In today's society,big data has become one of the most representative features of The Times,widely exists in all walks of life and life.Data mining and machine learning is an important part of big data technology.It is an important topic in the field of artificial intelligence to build a large-scale artificial intelligence model through mining and machine learning of big data information.Classification is one of the important research contents of big data mining.As a research branch of data mining,the essence of classification is to construct a classification model by learning the categories of training data,and to predict the categories of unknown samples according to classification rules.Based on neighborhood rough set and fuzzy rough set theory,this paper establishes different classification models and designs two classification algorithms.1.Rough neighborhood classification based on decision consistency.Firstly,according to the neighborhood rough set theory,the sample neighborhood is redefined,and based on the principle of decision consistency,the concepts of neighborhood purity are proposed.Then the selection method of sample neighborhood radius is discussed and analyzed to make it meet the decision consistency condition.According to the different distribution of samples,the method of selecting the neighborhood radius of samples is put forward.Design and determine the center sample of elementary class and its corresponding radius value algorithm;In the process of algorithm iteration,considering that the classification time should be shortened as much as possible,pruning algorithm was introduced to get the final sample of the class center and its corresponding radius,so as to complete the training and learning process.In order to predict the category of unknown samples,a neighborhood rough classifier based on decision consistency was constructed,and the relevant experiments of data set design in UCI were used for verification and analysis.2.Fuzzy rough classification of neighborhood based on decision fusion.In theclassical fuzzy rough set model,the sample decision is made by searching the whole sample.In fact,the decision of a sample is often related to its similar neighbor samples,so the concept of neighborhood is introduced into the classical fuzzy rough set model,and the fuzzy rough set of neighborhood is proposed.Considering that the selection of neighborhood threshold has different influence on fuzzy decision,the concept of threshold weight is introduced.Through the fusion of classification rules under different neighborhood thresholds,the fusion decision is closer to the real decision,and an optimization problem is constructed to minimize the difference between the fusion decision and the real decision.The problem of optimal weight of classification is transformed into an optimization problem with threshold weight as the variable,and then the fusion weight is obtained.Based on fusion decision,classification prediction of unknown samples was carried out,and relevant experiments of data set design in UCI were used for verification and analysis.