An Improved FCM Algorithm And Its Application Research

Data mining technology discovers valuable information and knowledge from the massive data.Cluster analysis,as an important branch in data mining technology,is widely used in biomedical,financial analysis,Internet,image analysis,and so on.Among the great deal of clustering approaches,the partition-based clustering algorithms have the advantages of simple idea and low time complexity,which is more suitable for processing the massive data,thereby has more research significances and practical values.Fuzzy C-means(FCM)clustering algorithm is a commonly used partition-based clustering method.But FCM has some disadvantages such as uncertain C value and sensitive center point of the initial cluster.In view of the above problems,this dissertation proposes an improved algorithm,and applies this algorithm to the cigarette quality inspection.The main work of this dissertation is as follows:(1)The Canopy algorithm and density peak clustering algorithm are applied to the FCM algorithm,and a new method to improve the selection of initial cluster center of the FCM algorithm is proposed.To solve the problem that the number of clusters in the traditional FCM algorithm is uncertain and the clustering result is too sensitive to the selection of the initial cluster center,this paper proposes an FCM algorithm that combines the Canopy algorithm and the density peak algorithm.Firstly,the Canopy algorithm was used to cluster the sample datasets,and the number of cluster center points was obtained.Based on the idea of density peak algorithm,the cluster center selection index is constructed to determine the more accurate initial cluster center point.The experimental results on the UCI dataset show that the proposed algorithm can find the optimal cluster center point faster than the traditional FCM algorithm,and can also accelerate the convergence speed.(2)The clustering algorithm proposed in this paper is applied in tobacco industry.Two sets of experiments are performed on the real cigarette data set.Firstly,using the clustering algorithm proposed in this paper to classify and predict the cigarette branches,the cigarettes of different brands are separated,and the cigarettes of the same brand are clustered into the same cluster.Secondly,by analyzing the physical indexes such as single weight,length,circumference and absorption resistance of tobacco,the clustering algorithm proposed in this paper is used to find out the outliers and the unqualified products of cigarette data,calculate the qualified rate of each brand.Two sets of experiments have proved that the clustering algorithm proposed in this paper can obtain better results under the division and prediction task of cigarettes,which can assist the cigarette factory in the quality inspection of cigarettes.