Smart k-means: An open-source tool for reliable representation of multidimensional phenomena

The representation of multidimensional phenomena requires clustering methods that are both reliable and interpretable. Traditional k-means, however, faces persistent difficulties in selecting the number of clusters and ensuring adequate cohesion and separation, which may compromise substantive interpretation. To address these limitations, this paper proposes a methodological framework that combines a user-defined interval for the number of clusters, a systematic search over parameter configurations, and iterative sub-indicator selection based on either exhaustive optimization or Shannon entropy, retaining only solutions that satisfy predefined quality thresholds. The study’s applied contribution is Smart k-means, an open-source Python implementation that operationalizes this framework and enables users to explore admissible clustering solutions and visualize the resulting groups. The proposed approach supports the identification of cluster structures with improved internal coherence, external separation, and interpretability in applications involving multidimensional phenomena.