Research On Classification Method Based On Fuzzy Rough Sets And Hypernetworks

Hypernetworks is a rule-based classification model,which can effectively deal with various classification problems.It has been widely used in pattern classification,machine learning,bioinformatics and other fields.In practical applications,the traditional hypernetworks model mainly has the following two problems:(1)The model can only deal with discrete data;(2)There is a large randomness in the process of hyperedge initialization.Fuzzy rough set can effectively deal with data with complex attributes.Therefore,this thesis proposes a fuzzy hypernetworks model based on the traditional hypernetwork and combined with the fuzzy rough set theory.The main contents of the work are as follows:(1)Combining with fuzzy rough set theory and the related knowledge of hypernetworks,a classification method based on fuzzy hypernetworks is proposed.Firstly,the ?-equivalent class sample set of each training sample is calculated according to the optimal fuzzy similarity threshold ?,and the samples in the training set are divided into boundary domain samples,positive domain samples and negative domain samples according to the class distribution of the set.The samples of different regions generate hyperedges according to different rules.Secondly,in the process of super-edge substitution,the fuzzy hypernetworks divides the hyperedges set into three regions according to the classification effect of hyperedge on the training samples: negative region,positive region and boundary region,and different regions set different substitution rules.Finally,when classifying,the fuzzy hypernetworks judges the class of sample according to the voting results of the ?-equivalence class hyperedges of the sample to be classified.In order to verify the performance of the algorithm,experiments are carried out on 15 UCI data sets,and the Accuracy,Precision and Recall are used as evaluation indexes to prove that the fuzzy hypernetworks has high applicability.It has a good classification effect on different data sets.(2)In order to solve the problems of long running time and low execution efficiency of serial algorithm in dealing with large-scale data sets,this thesis implements a parallel fuzzy hypernetworks classification algorithm based on Spark distributed parallel computing framework.Firstly,read the data from HDFS to create the train RDD and the test RDD,and convert the train RDD to get the hyperedges RDD;Then,the parallel hyperedge substitution algorithm is used for evolutionary learning;Finally,the output fuzzy hypernetworks model is used to classify the data in the test RDD.The experimental results show that the parallel algorithm not only maintains the effectiveness of the original serial algorithm,but also greatly reduces the running time.