The Rand index continues to be one of the most popular indices for assessing agreement between two partitions. The Rand index combines two sources of information, object pairs put together, and object pairs assigned to different clusters, in both partitions. Via a decomposition of the Rand index into four asymmetric indices, we show that in many situations object pairs that were assigned to different clusters have considerable impact on the value of the overall Rand index.
|Title of host publication||Advanced Studies in Classification and Data Science|
|Editors||Tadashi Imaizumi, Akinori Okada, Sadaaki Miyamoto, Fumitake Sakaori, Yoshiro Yamamoto, Maurizio Vichi|
|Number of pages||13|
|Publication status||Published - 2020|