Robust Clustering of Cylindrical Objects

Cylindrical data arise in many scientific domains and consist of observations combining one periodic angular variable with one or more linear variables. Clustering methods designed for Euclidean spaces fail to respect the periodic nature of angular variables, often resulting in arti-ficial boundaries and distorted cluster structures. Although geometry-aware extensions of centroid-based algorithms have been developed for cylindrical data, medoid-based approaches remain comparatively under-explored despite their advantages in interpretability and robustness.This work proposes a robust medoid-based clustering framework that integrates appropriate circular-linear dissimilarities with robust dispersion measures, yielding an effective method for clustering cylindrical data in the presence of outliers.