Combinatorial Fusion Analysis

Abstract

Combination methods have been investigated as a possible means to improve performance in multi-variable (multi-criterion or multi-objective) classification, prediction, learning, and optimization problems. In addition, information collected from multi-sensor or multi-source environment also often needs to be combined to produce more accurate information, to derive better estimation, or to make more knowledgeable decisions. In this chapter, we present a method, called Combinatorial Fusion Analysis (CFA), for analyzing combination and fusion of multiple scoring. CFA characterizes each Scoring system as having included a Score function, a Rank function, and a Rank/score function. Both rank combination and score combination are explored as to their combinatorial complexity and computational efficiency. Information derived from the scoring characteristics of each scoring system is used to perform system selection and to decide method combination. In particular, the rank/score graph defined by Hsu, Shapiro and Taksa (Hsu et al., 2002; Hsu & Taksa, 2005) is used to measure the diversity between scoring systems. We illustrate various applications of the framework using examples in information retrieval and biomedical informatics.