Search papers, labs, and topics across Lattice.
This paper reviews various external validity indexes for evaluating clustering when ground truth is available, emphasizing set-matching-based measures. It highlights the centroid index (CI) as an intuitive and explainable cluster-level measure, while also discussing point-level measures like the pair-set index (PSI) and clustering accuracy (ACC). The findings suggest that the choice of index should align with the evaluation goals, particularly regarding the importance of individual points versus overall cluster performance.
Choosing the right clustering evaluation index can significantly impact the interpretation of results, with the centroid index emerging as a standout for its intuitive clarity.
External indexes can be used for cluster evaluation when ground truth is available. We review the most common external validity indexes focusing on set-matching-based measures. We recommend centroid index (CI), because it is an intuitive cluster-level measure with an explainable result. If we need a more fine-tuned, point-level measure, there are more choices. Pair-set index (PSI) provides a normalized score which is not biased by cluster sizes. If all points should matter equally, then clustering accuracy (ACC) or any other set-matching measure is suitable.
