Influence of weight function for similarity measures

Abstract

The mainstream for dealing with pattern recognition problems is to develop new similarity measures, and then to compare outcomes among different measures. Along with a study trend focusing on developing new similarity measures for pattern recognition problems, this study tackles the issue of tuning weight functions of the existing measures. In this study, a detailed examination is executed to point out that a chosen weight function decides the pattern for a given example. The main contribution of the paper is to provide analytic derivations to explain the influence of weights for both discrete and continuous cases which supports our claims with mathematical foundations. With findings from this study, we expect a sensitivity analysis of the weights and exploring procedures in deciding a reasonable weight function for applications that can be set for future studies.

Keywords

• similarity measure
• intuitionistic fuzzy sets
• pattern recognition