Research And Application Of Granularity Clustering Algorithm For Mixed Attribute Data Under Dominance Relation

Cluster analysis can use a similarity measure to mine the valuable knowledge based on the characteristics and internal structure of data objects without prior knowledge.Most traditional clustering algorithms can only analyse the data consisting of a single type of attribute,such as numeric or category,and there are relatively fewer clustering algorithms for hybrid attribute data.However,most of the data in the real world has mixed attributes.Therefore,the research on clustering algorithms for hybrid data has important theoretical significance and practical value.Moreover,by using traditional crisp clustering algorithms,whether a data object belongs to a cluster or not is unambiguous.However,due to the objective laws of ambiguity in nature,the boundaries between clusters are not very clear in general,and many data in reality are located between the two partitions.Therefore,it is of great practical significance to explore the effective soft clustering algorithms for mixed attribute data.In this paper,our work includes qualitative combination algorithm for rough clustering based on fuzzy dominance relations,Qualitative combination scheme based on shadow rough clustering for mixed attribute data,and the application of qualitative combination scheme for rough clustering in grain post-harvest loss analysis.The main research work are as following:(1)Rough clustering qualitative combination algorithm based on fuzzy dominant relationship.In the real world,when customers select a product or service,they often focus on one of attributes after considering problems at different levels.In addition,the evaluations of attributes of products or services are usually shown by a sequence of data.However,most of the existing algorithms for mixed attribute data consider all attributes comprehensively,and classify clusters by similarity measurement,which can find the overall optimal cluster while fail to meet the requirements of customers for a particular attribute,and make the semantic structure analysis of class clusters difficult.In the qualitative combination scheme(QRD),different attribute particles are clustered separately to obtain the subclass cluster,which can not only represent the overall optimal cluster to customers,but also give consideration to the special requirements of customers on a certain attribute.At the same time,it is convenient to analyze the semantic structure of clusters by combining subclusters.However,this scheme is only suitable for the numerical attribute data set,which means that it cannot be applied to the cluster analysis of mixed attribute data.What's more,the result of this combination scheme is too ideal.In practical application,the number of data attributes is much more than two,which often leads to the problem of excessivefragmentation of the combined class cluster.In order to perform qualitative combinatorial clustering analysis for mixed data containing numerical and noun attributes and further describe order structure in detail,a qualitative combinatorial algorithm for rough clustering scheme is proposed based on fuzzy dominance relation,and the superiority of the algorithm is verified by the comparative analysis of examples.(2)Qualitative combination scheme of hybrid data based on shadow rough clustering.Qualitative combination algorithm for rough clustering based on fuzzy dominance relation has been applied in the mixed attribute data sets,for which the clusters are derived from the combination of subclusters based on individual attributes.However,it ignores the association relationship between information granules,which may decrease the overall accuracy of QRD.Therefore,the shadow set is adopted to mine the connection between information granules.In this paper,a qualitative combination scheme considering the level of granularity is put forward based on shadow rough clustering for mixed attribute data.Firstly,clustering the data set according to individual attributes.Then,based on the clustering of one attribute,the average membership of subclusters based on other attributes are computed.Finally,the results of all attributes iteration are combined.The qualitative combination algorithm on the basis of granularity level mines the relationship between each attribute granule,which reduces the information loss and improves the overall accuracy.Moreover,the effectiveness of the algorithm is validated by performing simulation and comparative analysis on the UCI data set.(3)The application of qualitative combination scheme for rough clustering in grain post-harvest loss analysis.Based on the above-mentioned research,qualitative combination scheme for rough clustering is particularly applied in grain post-harvest loss analysis.The post-harvest loss data is a typical mixed attribute data.To be more specific,a piece of data information can include various attributes such as raw material origin,production,storage,etc.,which are not all useless but directly or indirectly affects the loss rate.ithe application,post-harvest loss data based on every attribute is clustered one by one so as to explore the main factors affecting the post-harvest loss of grain,which can provide decision-making guidance for reducing post-harvest losses and improving profit margin.