Skeleton Clustering: Dimension-Free Density-Aided Clustering

We introduce a density-based clustering method called skeleton that can detect clusters in multivariate and even high-dimensional data with irregular shapes. To bypass the curse of dimensionality, we propose surrogate density measures are less dependent on dimension but have intuitive geometric interpretations. The framework constructs concise representation given as an intermediate step be thought combination prototype methods, clustering, hierarchical clustering. show by theoretical analysis empirical studies leads to reliable scenarios.