SSDP-H: Hybrid Approaches for Subgroup Discovery on High-Dimensional Datasets

This work presents new approaches to the task of Subgroup Discovery with a focus on high-dimensional databases: SSDPH-E and SSDPH-SD. They’re hybrid, top-k, and mono-objective approaches resulting from the integration of SSDP+ with exhaustive search and SD, respectively. SSDP+ tends to be better than SD in breadth-first search and worse in depth-first search. Additionally, SSDP+ is faster in runtime but inferior in subgroup quality compared to the exhaustive search. Therefore, we believe that a hybrid approach can reduce the limitations of these algorithms by integrating their strengths. We evaluated the impact of these hybridizations with respect to criteria such as quality, diversity and execution time. The experiments were performed on 10 databases, covering bioinformatics and relevant problems in Brazil. The results indicate that the proposed solutions may represent a relevant advance for the Subgroup Discovery area.