A greedy and optimistic clustering for leveraging individual covariate uncertainty

Abstract

In this study, we examine a clustering problem where each individual element in a dataset has covariates associated with element-specific uncertainty. More specifically, we consider a clustering approach that preliminarily applies a non-linear transformation to the covariates, to capture the hidden data structure; we empirically approximate the sets representing the propagated uncertainty for the pre-processed features and propose a greedy and optimistic clustering (GOC) algorithm. This algorithm identifies better feature candidates within these sets, resulting in more condensed clusters. As a key application, we apply the GOC algorithm to synthetic datasets of the orbital properties of stars, generated through our numerical simulations that mimic the formation process of the Milky Way. The GOC algorithm demonstrates improved performance in identifying sibling stars originating from the same dwarf galaxy. These realistic datasets are also publicly available at https://github.com/oknakfm/GOC.