Parameter Free Clustering Algorithm Based on Density and Natural Nearest Neighbor

The purpose of clustering algorithm is to explore the correlation of a large number of unlabeled data, so as to find valuable information in chaotic data. There are many kinds of clustering algorithms, but most of them need one or more parameters to be selected based on experience, and the selection of parameters will directly affect the accuracy of clustering algorithm, for example, DBSCAN needs domain radius and number of radius points; k-means algorithm needs to know the number of clusters in advance, k-nearest neighbor algorithm needs to be selected. Choose the appropriate number of neighbors, etc. In order to realize parametric clustering, we adopt the concept of natural nearest neighbor, and let data points discover neighbors independently by iteration. To make up for the problem that the selected K value may not be appropriate, in the process of clustering, we proposes a method of similarity between clusters, which is used to modify the misclassified clusters. Finally, the similarity of observation points is proposed. To distinguish boundary points from outliers. By comparing with DBSCAN, BIRCH and K-MEANS, proved that our algorithm can achieve good performance than other algorithms.