Building upon the ClicoT algorithm’s concept hierarchy approach for mixed-type data clustering, we propose WP-ClicoT (Weighted Partition ClicoT), a novel weighted partition-based algorithm that dynamically assigns and calculates weights for categorical data types while establishing quantifiable associations between continuous and categorical attributes. This graduated weighting represents a fundamental shift from binary decisions to continuous relevance modelling, enabling a more nuanced representation of categorical attribute importance across clusters. Unlike ClicoT’s binary specific/non-specific element selection, WP-ClicoT introduces a continuous weighting scheme that captures varying degrees of relevance for categorical attributes. Our approach employs mutual information-based weight calculation for categorical features and introduces a novel association metric that bridges the gap between heterogeneous data types. Through a partition-based framework with adaptive weight refinement, WP-ClicoT achieves superior clustering performance while maintaining interpretability. Extensive experiments on synthetic and real-world datasets demonstrate that WP-ClicoT outperforms ClicoT and other state-of-the-art algorithms, with NMI improvements of up to 18% on datasets with complex categorical structures. We also provide comprehensive comparisons with deep learning-based approaches and demonstrate competitive performance with superior interpretability compared to existing methods.
Dutt et al. (2026) studied this question.