Hybrid MCDM and Explainable Clustering for Sustainability Performance Segmentation of MSMEs
Аннотация
The heterogeneity of Micro, Small, and Medium Enterprises (MSMEs) challenges the design of targeted sustainability policies, as uniform interventions fail to address structurally distinct enterprise needs. This study proposes a Hybrid Decision Framework (HDF) combining Multi-Criteria Decision-Making with explainable machine learning for data-driven sustainability segmentation. Rather than deriving new weights, the framework adopts previously derived BWM-Entropy-D Numbers (BED) weights across eight dimensions within an unsupervised UMAP–DBSCAN pipeline, with Decision Tree and Random Forest models extracting interpretable rules and criterion importance. Applied to 100 MSMEs, the framework identified four clusters and an atypical group of nine enterprises (Silhouette = 0.450, Davies–Bouldin = 0.611, Calinski–Harabasz = 229.9 on the 91 clustered observations). BED weighting altered the segmentation relative to an unweighted configuration (ARI = 0.577). When K-Means, Agglomerative, and GMM were fixed at the same number of clusters ($$\varvec{k = 4}$$), they obtained higher Silhouette and Calinski–Harabasz values in the UMAP space yet recovered a largely similar partition (ARI up to 0.80), indicating that the segment structure is not algorithm-specific; UMAP–DBSCAN was retained because it determines the number of segments without prior specification and isolates atypical enterprises instead of forcing every case into a predefined tier. Random Forest analysis identified economic performance and innovation/adaptability, rather than the highest-weighted criterion, as the strongest empirical discriminators, revealing a divergence between normative and data-driven importance. A proposed five-category policy framework aligns intervention types and intensities with each segment’s relative profile, offering development agencies and financial institutions a transparent basis for prioritising limited resources.