An adaptive clustering algorithm based on boundary detection

To solve the problem that the partition clustering algorithm usually needs to specify the number of clusters artificially and the poor clustering effect on nonconvex datasets, this paper proposes an adaptive clustering algorithm based on boundary detection (BAC). BAC algorithm searches the boundary points of each cluster according to the global distribution characteristics of all samples in the sample space, then connects the boundary points to form the boundary shape closure of the cluster, and finally propagates the cluster labels of the boundary points to the non boundary points according to the nearest neighbor principle. BAC algorithm finds the number of clusters adaptively by searching the number of boundary shapes in the sample space, and it is not sensitive to the shape of clusters, so it can detect clusters with complex shapes. Experimental evaluation is carried out on a large number of datasets, and compared with K-means, K-means++ and DPC algorithms. The results show that the clustering performance of BAC algorithm is better than other algorithms.