Monotone Clusters With Ordinal Interpretation

ABSTRACT There has been a growing demand in various sectors, such as healthcare, finance, and social sciences, for clustering methods that not only find quality clusters in data but also provide inherent order among the clusters for better decision‐making and risk assessment. Traditional clustering methods, though effective at grouping data, often fall short in delivering interpretable and structured clusters. This paper introduces Sparse MOnotone CLustering (SMOCL), a novel unsupervised algorithm specifically designed to address these challenges by focusing on ordinal interpretability. SMOCL integrates generalized additive models with clustering techniques in an iterative manner. A key benefit of SMOCL is its ability to select variables that are monotonically related to the identified cluster labels, which greatly enhances the interpretability of the clustering results. The algorithm's effectiveness is demonstrated through both synthetic and real‐world data sets.