LLAhclust: Hierarchical clustering of variables or objects based on the likelihood linkage analysis method

The likelihood linkage analysis is a general agglomerative hierarchical clustering method developed in France by Lerman in a long series of research articles and books. Initially proposed in the framework of variable clustering, it has been progressively extended to allow the clustering of very general object descriptions. The approach mainly consists in replacing the value of the estimated similarity coefficient by the probability of finding a lower value under the hypothesis of 'absence of link'. The package LLAhclust contains routines for computing various types of probablistic similarity coefficients between variables or object descriptions. Once the similarity values between variables/objects are computed, a hierarchical clustering can be performed using several probabilistic and non-probabilistic aggregation criteria, and indices measuring the quality of the partitions compatible with the resulting hierarchy can be computed.

Downloads: