Discovering informative subgroup sets is a core goal in subgroup discovery; however, it remains challenging. Existing methods often require manual, problem-specific parameter tuning, which limits scalability and reproducibility. Many methods also optimize subgroup quality using a single metric, assuming it alone can capture relevance. This is restrictive because subgroup quality is inherently multidimensional, involving trade-offs among accuracy, coverage, and diversity. In addition, candidate subgroups are usually evaluated in isolation, without considering their overlap or overall contribution to the final set. To overcome these issues, we propose DASSD (Dynamic and Adaptive Subgroup Set Discovery), a heuristic algorithm that discovers high-quality, non-redundant subgroup sets. DASSD jointly optimizes global set quality and redundancy using a dual-metric strategy based on information gain and odds ratio. A Minimum Redundancy Maximum Relevance (mRMR) pruning mechanism guides the search toward informative and diverse subgroups. The algorithm also uses a data-driven dynamic threshold to adapt parameters automatically, removing the need for manual tuning and enhancing robustness across datasets. Experiments on multiple benchmark datasets show that DASSD discovers high-quality subgroup sets with minimal redundancy and outperforms state-of-the-art methods under multiobjective evaluation, achieving a roughly 200% improvement in mRMR, while remaining computationally efficient.
García et al. (Wed,) studied this question.