funBI: a Biclustering Algorithm for Functional Data
Tuesday 3rd December 2019
Di Iorio, J.; Vantini, S.
In order to group objects, a wide literature of methods, the majority of them known as clustering and biclustering methods, was created. In the meanwhile, the scientific community tried to defy the curse of dimen- sionality, dealing with problems characterized by data with one infinite continuous dimension: functional data. Even if many old and new clus- tering algorithms were generalized to these new types of data, biclustering methods did not share the same destiny. This paper fills the literature gap by defining the concept of bicluster for data described as a set of functions, and by introducing funBI, the first biclustering algorithm that permits to find functional biclusters, i.e. subsets of functions that exhibit similar behaviour across the same continuous subsets of the domain. funBI is a three-step algorithm based on DIANA, the most famous divisive hierar- chical clustering method. The use of DIANA allows to visualize and to guide the searching procedure using dendrograms and cutting thresholds. Biclustering Clustering Functional data