Nearest Prototype Classification of Special School Families Based on Hierarchical Compact Sets Clustering

The family orientation process in Cuban Schools for children with Affective – Behavioral Maladies (SABM) involves clustering and classification of mixed type data with non-symmetric similarity functions. To improve this process, this paper includes some novel characteristics in clustering and prototype selection. The proposed approach uses a hierarchical clustering based on compact sets, making it suitable for dealing with non-symmetric similarity functions, as well as with mixed and incomplete data. The proposal obtains very good results on the SABM data, and over repository databases. In addition, the proposed clustering method is able to detect the true partitions of data and it was significantly better with respect to others according to external validity indexes. In prototype selection, the proposal obtains a highly reduced prototype set, while maintains the original classifier accuracy.