An adaptive density clustering approach with multi-granularity fusion

The real-world dataset exhibits diversity, incorporating instances with complex shapes and significant differences in density hierarchy, potentially disrupted by noise. However, most clustering algorithms typically rely on single-granularity fusion, requiring the pre-setting of global parameters for the entire dataset. Nevertheless, these global parameters may not adequately adapt to clusters with varying hierarchies or shapes, consequently reducing the clustering effectiveness. Therefore, we propose an adaptive density clustering approach with multi-granularity fusion. This approach characterizes the dataset with multi-granularity, forming natural granular-ball. After processing these natural granular-ball, overlapping ones are fused to yield the final clustering result. The entire approach adeptly identifies datasets with significant differences in shape or density hierarchy and exhibits a certain level of robustness. All codes have been released at https://github.com/xjnine/NGBC.