Comparative Analysis of Cluster Interpretation Methods via Synthetic Datasets

This study evaluates Random Forest surrogates, LIME, and PCA for detecting structured patterns in clustering results. Using synthetic datasets with injected patterns, the researchers demonstrate that existing post-hoc methods struggle to identify complex cluster structures compared to simple feature importance.