Abstract: Aiming at the problem that the traditional Yinyang algorithm did not use the data structure, resulting in low computational efficiency, we proposed an efficient Yinyang k-means clustering algorithm. The algorithm decomposed the original data layer by layer according to the data similarity, and established a full m-tree structure to store the data of each layer. The weighted data was established based on the data information stored in each leaf node of the tree structure, and the weighted Yinyang k-means algorithm was run to obtain the convergence center. In the original data, the convergence centers of the weighted data were taken as the initial condition to run the traditional Yinyang k-means algorithm to further optimize the objective function value. Comparative experiments with k-means, traditional Yinyang k-means and other two acceleration algorithms on five UCI data sets show that the proposed algorithm has a high acceleration ratio, and the solution accuracy is basically equivalent to Yinyang k-means clustering.

Key words: cluster analysis, Yinyang k-means algorithm, k-means , algorithm, data weighting