Although Hierarchical Decomposition Methods (HDMs) are crucial for simplifying complex multi-class problems, most of them lack a mechanism for optimally structuring the class space based on intrinsic data properties. To overcome this issue, we present the Mental Image Decomposition ( MID ) framework, a new HDM that uses the information content of DRASiW “Mental Images” to create an optimal hierarchical tree aligned with data. We rigorously validated the MID method through extensive experiments on 26 standard datasets. We compared its performance against flat classification approaches and 12 diverse machine learning (ML) classifiers, as well as seven established HDMs. Our comprehensive analysis confirmed that MID is the leading method, with results demonstrating that it secured the global ranking of 1 across all comparative performance and stability metrics ( F1-score and accuracy gain) compared with all competing HDMs. Furthermore, we showed that MID significantly improves ensemble classifiers such as AdaBoost and provides the optimal balance between performance and interpretability when integrated with the r DAB r system (a variant of the DRASiW architecture). This work establishes MID as a highly effective and robust contribution in the field of hierarchical classification. • A novel Hierarchical Decomposition Method (HDM) driven by DRASiW “Mental Image” is proposed: Mental Image Decomposition ( MID ). • MID guides nested dichotomy construction by maximizing inter-group separability and preserving intra-group class cohesion. • Extensive experiments on 26 datasets show MID consistently outperforms 12 classifiers and 7 HDMs.
Sorgente et al. (Sun,) studied this question.