Method for clustering non-spherical data by using IK-means algorithm

Abstract

The invention discloses a method for clustering non-spherical data by using an IK-means algorithm, and the method comprises the steps: carrying out the initial clustering of an original data set through employing a classical K-means algorithm, and obtaining a Ksub sub-cluster; and calculating the connectivity among the sub-clusters, and merging the sub-clusters of which the connectivity is greater than a connectivity threshold to obtain a final clustering result of the data set. According to the method, data of any shape can be clustered, the clustering effect is good, and particularly, the method has more advantages in clustering of non-spherical data.

Description

technical field [0001] The invention relates to a method for realizing clustering of non-spherical data by using IK-means algorithm. Background technique [0002] The rapid development of Internet technology has generated a large amount of data. To analyze and process these data in a timely manner, some tools are needed. Clustering algorithm is one of the important ones. It is an unsupervised learning method that can be processed for data mining. There are many clustering algorithms, which can be roughly divided into several categories according to their working mechanism: layer-based, such as BIRCH algorithm, CURE algorithm; grid-based, such as STING algorithm, CLIQUE algorithm; partition-based, such as K-means algorithm; Based on graph theory, such as spectral clustering algorithm; density-based, such as DBSCAN algorithm, OPTICS algorithm; model-based, such as Gaussian mixture model (GMM) algorithm; fuzzy-based clustering, such as FCM algorithm. Each clustering algorit...

AI Technical Summary

Problems solved by technology

And those algorithms that can cluster non-spherical data, such as DBSCAN, spectral clustering, but they have some shortcomings in common: high computational cost, insurmountable difficulties in selecting initial parameters; high time complexity, for massive data The analysis is relatively slow, and it is difficult to meet the real-time requirements

However, the model selection of the kernel function is a difficult problem, and different kernel functions have different clustering results.

In addition, the kernel function increases the calculation cost and reduces the clustering efficiency

It cannot afford the timely processing of massive data streams generated in the current Internet era

Images