Research output: Contribution to Journal/Magazine › Journal article › peer-review
Research output: Contribution to Journal/Magazine › Journal article › peer-review
}
TY - JOUR
T1 - Density-based averaging - a new operator for data fusion
AU - Angelov, Plamen
AU - Yager, Ronald
PY - 2013/2/10
Y1 - 2013/2/10
N2 - A new data fusion operator based on averaging that is weighted by the density of each particular data sample is introduced in this paper. The proposed approach differs from other weighted averages by its suitability to on-line, real-time applications due to the fact that recursive calculations are being used. It alsodiffers by the fact that it is non-parametric. The proposed operator has a very wide area of possible applications same as the traditional average and most of the other weighted averages. This includes, but is not limited to clustering, classification, pattern recognition, group decision making approaches, datafusion, etc. Some illustrative numerical examples are provided mainly as a proof of concept, including its application to classification. Two simple, yet very effective classification approaches based on the density-based weights called ‘one-rule-per-class’ or 1R/C and on the minimum distance to weighted class mean has been introduced. Further work will focus on more application-oriented studies that cover various practical applications to clustering and use of different distance measures.
AB - A new data fusion operator based on averaging that is weighted by the density of each particular data sample is introduced in this paper. The proposed approach differs from other weighted averages by its suitability to on-line, real-time applications due to the fact that recursive calculations are being used. It alsodiffers by the fact that it is non-parametric. The proposed operator has a very wide area of possible applications same as the traditional average and most of the other weighted averages. This includes, but is not limited to clustering, classification, pattern recognition, group decision making approaches, datafusion, etc. Some illustrative numerical examples are provided mainly as a proof of concept, including its application to classification. Two simple, yet very effective classification approaches based on the density-based weights called ‘one-rule-per-class’ or 1R/C and on the minimum distance to weighted class mean has been introduced. Further work will focus on more application-oriented studies that cover various practical applications to clustering and use of different distance measures.
KW - weighted averages
KW - data fusion
KW - data density
KW - Cauchy kernel
KW - Fuzzy classifiers
U2 - 10.1016/j.ins.2012.08.006
DO - 10.1016/j.ins.2012.08.006
M3 - Journal article
VL - 222
SP - 163
EP - 174
JO - Information Sciences
JF - Information Sciences
SN - 0020-0255
ER -