A new similarity measure for objects that are represented by feature vectors of fixed dimension is introduced. It can simultaneously deal with numeric and symbolic features. Also, it can tolerate missing feature values. The similarity measure between two objects is described in terms of the similarity of their features. IF-THEN rules are being used to model the individual contribution of each feature to the global similarity measure between a pair of objects. The proposed similarity measure is based on fuzzy sets and this allows us to deal with vague, uncertain and distorted information in a natural way. Several formal properties of the proposed similarity measure are derived; in particular, we show that the measure can be used to model the Euclidean distance as well as other, non-Euclidean distance functions. Also, an application of the proposed similarity measure to nearest-neighbor classification in a medical expert system is described.
@article{urn:eudml:doc:39216,
title = {Rule-based fuzzy object similarity.},
journal = {Mathware and Soft Computing},
volume = {8},
year = {2001},
pages = {113-128},
zbl = {1014.68162},
mrnumber = {MR1864068},
language = {en},
url = {http://dml.mathdoc.fr/item/urn:eudml:doc:39216}
}
Bunke, Horst; Fábregas, Xavier; Kandel, Abraham. Rule-based fuzzy object similarity.. Mathware and Soft Computing, Tome 8 (2001) pp. 113-128. http://gdmltest.u-ga.fr/item/urn:eudml:doc:39216/