SIMILARITY MEASURES FOR NOMINAL VARIABLE CLUSTERING

The paper deals with selected similarity measures which can be used for hierarchical clustering of nominal variables. These variables are commonly used in questionnaire surveys. Cluster analysis can be applied in case a reduction of a dataset size is welcomed. In this paper, there are examined several similarity measures for nominal variable clustering, which have been introduced in recent years. On the contrary to the simple matching coefficient, which is considered to be a basic similarity measure, they take into account more characteristics regarding the dataset, such as distribution of frequencies of categories. Therefore, they should provide better results in a comparison to the simple matching coefficient. The performance of clustering with selected similarity measures is examined on two real datasets. For cluster quality evaluation, indices based on the within-cluster variability have been chosen. All computations have been performed in the statistical systems Matlab, IBM SPSS Statistics and MS Excel.