Minimum variance and range-based k-means optimization algorithm-Rdk-means algorithm

To address the problem that the traditional k-means randomly generates initialized centroids leading to the instability of the algorithm, Rdk-means (Range and density k-means) is proposed to improve the clustering algorithm. The algorithm introduces the concept of global density to select sample points. The centroids are initialized by selecting the sample points with the maximum density and distributed by a certain distance. At the same time, the concept of spatial location is introduced in the process of sample partitioning to filter out the centroids that are closer to each sample point, thus avoiding unnecessary centroid distance calculation and optimizing the shortage of longer clustering time caused by the initialization centroid process. Through experimental validation and analysis, the algorithm has been improved in clustering effect and stability.