Research And Application Of The Clustering Analysis Based On Improved Multi-objective Immune Algorithm

With the advent of the era of big data,data is showing an exponential growth trend.At the same time,the demand for data analysis and mining is also surging.As an important tool of data mining,clustering analysis has always been popular in research,but each clustering algorithm has some defects.We need to optimize the clustering algorithm to improve the clustering accuracy.In this paper,two widely used clustering methods,intuitionistic fuzzy C-means clustering(IFCM)and spectral clustering(SC),are selected as the main research contents,and multi-objective immune algorithm(MOIA)is selected to optimize them.The main research work is as follows:(1)A grid-based multi-objective immunity algorithm(GMOIA)is proposed.Firstly,the algorithm uses a grid-based active antibody selection strategy to improve the uniformity of the non-dominated solution distribution.Secondly,a hybrid differential evolution strategy and adaptive mutation operator are designed to enhance the diversity of the population and help the algorithm jump out of the local optimum.Finally,eight commonly used benchmark problems are selected for experiment.Experimental results show that,compared with other four multi-objective genetic algorithms,the non-dominated solution set obtained by GMOIA is closer to the true Pareto-optimal Front(PF)and has a more uniform distribution.(2)A kernel-based intuitionistic fuzzy C-means clustering using GMOIA(KIFCM-GMOIA)is proposed.Firstly,for linearly inseparable data,it is generally difficult to obtain satisfactory results using an intuitionistic fuzzy C-means clustering algorithm with Euclidean distance,so the algorithm introduces the gaussian kernel function to calculate the similarity between data points.Secondly,the inter-cluster separation is introduced as the second objective function,and the grid-based multi-objective immune algorithm is used to find the optimal cluster centers,which improves the defect that the traditional fuzzy clustering algorithm is sensitive to the initial cluster centers.Finally,KIFCM-GMOIA is compared with three well-known fuzzy C-means clustering algorithms on 12 UCI datasets,which proves the effectiveness of the algorithm.(3)An automatic spectral clustering algorithm based on GMOIA(ASC-GMOIA)is proposed.First of all,a new antibody representation method is used to solve the shortcoming of the traditional spectral clustering algorithm that needs to specify the number of clusters in advance.In ASC-GMOIA,each antibody contains the activation threshold and the cluster center.Secondly,the two objective functions of intra-cluster compactness and inter-cluster overall variance are selected,and the improved multi-objective immune algorithm is used to determine the number of clusters and find the optimal cluster centers.Finally,ASC-GMOIA is compared with other automatic clustering algorithms and traditional spectral clustering algorithm on four common artificial datasets and four UCI datasets.The experimental results show that the ASC-GMOIA algorithm can automatically determine the number of clusters and improve the accuracy of clustering.(4)The proposed KIFCM-GMOIA is applied to brain MRI segmentation.First,the new weighted image is obtained using adaptive weighting method,and then KIFCM-GMOIA is used to guide image segmentation.The brain MR images used in the experiment included 2 artificial images and 6 real images.The final experimental results show that the KIFCM-GMOIA algorithm has obvious advantages over traditional image segmentation algorithms.