Related Research About Hybrid Attribute Reduction Based On Fuzzy Rough Set

With the rapid development of information technology and the explosive growth of various data sets,how to effectively mine information from data with uncertainty has become an important research topic.The major branches of traditional machine learning research are closely related to the attributes(or features)of the original data.Attributes in a database that are redundant for a classification task may make the machine learning task more complex or even degrade the learning performance.Accurate and interpretable selection of data attributes has become an important development direction.Rough sets,as a data-driven intelligent computational tool,can mine various types of data without a priori knowledge,and fuzzy sets have obvious advantages in dealing with uncertain information systems.Based on fuzzy rough set theory,this paper investigates a class of entropy-weighted fuzzy metrics and their induced weighted fuzzy neighbourhood rough sets and corresponding reduction algorithms,interval-type similarity relations and their induced interval-valued fuzzy rough sets and reduction algorithms,mixed data based on the idea of multiple granularity,and the combination of fuzzy rough sets and intelligent optimisation algorithms,respectively.The main research results and innovations of the thesis are as follows:(1)For the problem of reduction of real-valued and fuzzy-valued information systems,a similarity relation induced by an entropy-weighted fuzzy distance metric is constructed considering the inadequacy of Euclidean distance,a formula for the relative relation change of a single increase or decrease of attributes is derived which can effectively reduce the computational effort,the Lukaszewicz implication relation is introduced to construct a fuzzy rough set lower approximation;the monotonicity and closure of the corresponding degree function are proved.The corresponding heuristic attribute reduction algorithm is then designed and verified by examples based on the relevant UCI dataset,and the experimental results show that the present algorithm is well adapted to the corresponding mixed data.(2)For interval-valued fuzzy information systems,fuzzy decisions on the number of intervals are introduced by combining interval-valued ordinal relations and variable-precision compatibility relations.And the respective rough set models are induced for information systems with ordinal relations and general information systems respectively,incorporating the well-defined variable-precision compatibility relations into the fuzzy rough set models,thus reflecting more information of the original data in the attribute reduction and ensuring the adaptiveness of the model to the data.Finally,the concepts of fuzzy positive domain and dependence of intervalvalued fuzzy rough sets are induced,and a reduction algorithm for fuzzy intervalvalued data is proposed and experimentally verified and analysed.(3)For more general mixed data information systems,a novel weighted multigrain rough set model is proposed in conjunction with the idea of multi-granularity rough sets,and a formalized reduction algorithm is given.Due to the complexity of mixed data sets and the diversity of parameters,the reduction results may fluctuate drastically with the change of parameters,and sometimes there is the problem of repeated computation of some attributes resulting in reduced operational efficiency.The highly complementary nature of intelligent optimisation algorithms and rough sets is analysed,and a new adaptive reduction algorithm for fuzzy rough sets of mixed data is constructed in combination with genetic algorithms.An example validation is carried out in conjunction with some public datasets.