Granular Box Regression

Granular computing (GrC) has gained increasing attention in the past decade. Although not uniquely defined, its basic idea is to approximate detailed machine-like information by a coarser presentation on a human-like level. Within granular computing, the mapping of continuous variables into intervals plays an important role. These intervals are often prerequisites for the formulation of linguistic variables. In this paper, we suggest a piecewise interval approximation and propose granular box regression. Its objective is to establish relationships between independent and dependent variables by multidimensional boxes. We interpret granular box regression as interval regression and show its potential for the extraction of fuzzy rules from data. In two experiments, we apply granular box regression to an artificial as well as to a real dataset in the field of finance and evaluate its properties.