The Research On Adaptive Fuzzy Joint Points Clustering Algorithm

Fuzzy Joint Points(FJP)is a new type of fuzzy clustering algorithm which has emerged in recent years.The advantage is that there is no need to preset the number of clusters,and it Can judge the similarity between categories of data points through the connection relationship of data points,with good robustness.However,the algorithm also has several shortcomings in the application:(1)The FJP algorithm takes the number of clusters with the highest frequency in the result as the best cluster number.This method is not applicable in many high-dimensional data sets,and has certain randomness,which affects the accuracy of the algorithm.(2)The range of α-level attenuation is too large,resulting in many iterations of the algorithm,and the calculation time on a large data set will be long.(3)The FJP algorithm uses a simple Euclidean distance formula to calculate the fuzzy similarity between data elements,which has the result of distortion on the multi-dimensional complex data set.In response to the above problems,this dissertation makes the following two improvements:(1)To determine the optimal clustering number of the original FJP algorithm,the Gaussian radial kernel function of the K-interpolation simplex method is used to fully exploit the similarity between data elements,and the Kernels-VCN index is proposed to evaluate all the classification results.The internal data is as similar as possible and the data between the classes is as dissimilar as possible.Finally,the optimal index level and the optimal cluster number are automatically determined by the intuitive index values to improve the adaptability of the algorithm.(2)The α-division horizontal attenuation range for the FJP algorithm is too large,causing more iterations.The decision graph of the 2014《Science》density peak clustering algorithm is introduced to assist the FJP to determine the cluster center,and the attenuation range of the FJP optimal division level is reduced.,to achieve the purpose of reducing the amount of calculation of the FJP algorithm.Through experimental verification in UCI dataset and manual dataset,it can be concluded that the improved measures proposed in this paper can effectively improve the accuracy and computational efficiency of FJP algorithm.