Research And Application Of Classification Algorithms Based On Fuzzy Rules In Big Data

Fuzzy rule classifiers can effectively overcome the shortcomings of symbol rules,and is easier to express and understand.It has become a hotspot of classification algorithms.In big data environment,rich data sources make the data expression more complex and ambiguous,high dimension and sparse data aggravate the problem of "Dimension Calamity ",efficiency and accuracy of fuzzy rule classifier.This thesis studies fuzzy rule classifiers which can solve this problem well.Based on the idea eliminated the “Dimension Calamity” of the fuzzy classifers with fixed number of fuzzy rules(FCFFR),a fuzzy classifier with limited number of fuzzy rules(FCLFR)and a classifier model with limited number of fuzzy rules and inverse convolutional neural network(CMLFRICNN)was proposed to avoid disadvantages and make fully use of fuzzy rules.The main work of this thesis is as follows:1.Owing to the fact that lacking of efficiency in the existing fuzzy rule classifiers wasting a lot of time that people can not bear to establish the model,the FCLFR is proposed.The classifier adds basic rules to the warehouse including positive and negative rules,and at the same time,it blurs the process when the cost function is getting minimize value to achieve the purpose to improve efficiency.Compared to the FCFFR and others popular classification models,it shows that the FCLFR improves the efficiency obviously and the accuracy lightly in big data.2.On the basis of the first study,the model CMLFRICNN is presented to solve it.The CMLFRICNN is consist of two branches: the limited fuzzy rules branch(LFRB)and the fuzzy inverse convolution branch(FICB).This thesis also propose a simple feature selection algorithm(SFS)based on fuzzy rules to classify the samples set into the main features set and the candidate features set.The main features set is sent to the FICB to obtain the inverse convolution result,and the candidate feature set is sent to the LFRB to compute another result.Combing the two results,the CMLFRICNN could get the final classification result.Conpared with the FCFFR and others popular classification models,it indicates the validity of the CMLFRICNN.3.To design and implement a module called an application of CMLFRICNN algorithm.The application uses the CMLFRICNN model and the XGBoost classifier to classify the users who will be lost for various reasons in the next month.Finally,compared the validity of the two classifers,the CMLFRICNN get better.The research illustrates that the study of fuzzy rule classifiers with big data has improved the speed and efficiency of the model establishment,and has also improved the classification accuracy,which has certain practical value and research significance.