A multilevel neighborhood sequential decision approach of three-way granular computing

Abstract

The fusion of three-way decision and granular computing provides powerful ideas and methods to understand and solve the problems of cognitive science by thinking and information processing in threes. As a typical representation of three-way granular computing, sequential three-way decision focuses on making a multiple stages of decisions by a sequence of trisecting-acting-outcome (TAO) models. To construct more general granules, levels, and hierarchies, we investigate an integrative multi-granularity approach to sequential three-way decision in a neighborhood system by the evolution mechanism of data and parameters. We employ the γ-cut similarity neighborhood relation based on Gaussian kernel function to the hierarchical granulation of universe. Subsequently, we propose the multilevel neighborhood granular structures by the combinations of horizontal granularity and vertical granularity, and discuss the monotonicity of level measurements associated with the uncertainty of decision. Based on such a neighborhood structured approach, a multilevel framework of sequential three-way decision is examined from coarser to finer concerning the granularity of neighborhood information. Finally, we report a series of experiments to demonstrate the performance of proposed models and algorithms.