On a weighted ordered weighted averaging based similarity classifier

Abstract

The classification accuracy of a classifier depends largely on the aggregation operator it uses in fusion of information. This paper presents a similarity classifier that utilizes a weighted ordered weighted averaging (WOWA) operator. The applicability of the WOWA operator for aggregation of similarities is studied and tested with several Regular Increasing Monotonic (RIM) type weight generators. The RIM quantifiers used here are of the basic, polynomial, trigonometric, exponential, and of the logarithmic type. The proposed approach is tested on three real world data sets. Results obtained with the new method are compared with those obtained using two previously presented similarity classifiers. The proposed new classifier shows better performance on the data sets studied, which indicates that benefits can be gained from using a WOWA operator in the aggregation.