Interclass analysis in symbolic pattern classification problem (Q880914)

A Cartesian system model (CSM) is considered for description of data of mixed type, including interval valued, nominal and ordinal features. The authors consider two-class classification problems and define the notions of neighbourhood, inside view, outside view and generality. An algorithm is proposed for feature selection which searches for a minimal feature set which maximizes the size of outside view. Such algorithms can be used in pre-processing before classification. Applications to Iris data and Ionosphere data are considered.