An evaluation of the IFCS Cluster Benchmarking Data Analysis Challenge

Eight clusterings of a lower back pain dataset were submitted to the IFCS Benchmarking Cluster Analysis Challenge. The aim of the challenge was to find clusterings of the 112 baseline variables that help with predicting 9 outcome variables. These clusterings are compared here, using data visualisation (multidimensional scaling and discriminant coordinates on both baseline and outcome variables), outcome means and uncertainty intervals, and four cluster validation indices, namely the Average SilhouetteWidth, the Pearson correlation version of Hubert’s $\Gamma$, the Calinski/Harabasz index, and the Adjusted Rand Index. The different comparison approaches give quite different assessments of the clustering quality.