Kernel KMeans clustering splits for end-to-end unsupervised decision trees
This addresses the problem of interpretable clustering for small datasets, offering an incremental improvement over existing unsupervised tree methods.
The paper tackles the challenge of end-to-end unsupervised tree construction for clustering by introducing Kauri, a method that greedily maximizes the kernel KMeans objective without centroids. It shows that Kauri performs identically to recent methods with a linear kernel and often outperforms kernel KMeans combined with CART for other kernels.
Trees are convenient models for obtaining explainable predictions on relatively small datasets. Although there are many proposals for the end-to-end construction of such trees in supervised learning, learning a tree end-to-end for clustering without labels remains an open challenge. As most works focus on interpreting with trees the result of another clustering algorithm, we present here a novel end-to-end trained unsupervised binary tree for clustering: Kauri. This method performs a greedy maximisation of the kernel KMeans objective without requiring the definition of centroids. We compare this model on multiple datasets with recent unsupervised trees and show that Kauri performs identically when using a linear kernel. For other kernels, Kauri often outperforms the concatenation of kernel KMeans and a CART decision tree.