The impact of isolation kernel on agglomerative hierarchical clustering algorithms

Agglomerative hierarchical clustering (AHC) is one of the popular approaches. AHC generates a dendrogram that provides richer information and insights from dataset than partitioning clustering. However, major problem with existing distance-based methods is: it fails to effectively identify adjacent clusters varied densities, regardless cluster extraction applied resultant dendrogram. This paper aims reveal root cause this issue solution by using data-dependent kernel. We analyse condition under which fail extract clusters, give reason why kernel an effective remedy. leads new approach kernerlise algorithms including traditional algorithms, HDBSCAN, GDL, PHA HC-OT. Our extensive empirical evaluation shows recently introduced Isolation Kernel produces higher quality or purer distance, Gaussian adaptive in all above mentioned algorithms.