Semi-supervised fuzzy clustering algorithm based on prior membership degree matrix with expert preference

Existing pre-processing methods for the prior membership degree matrix suffer from the following issues: (1) The labeling constraints for prior membership degree matrix have an effect on the expert’s judgment on the prior membership degree, which easily causes the distortion problem of the prior mem...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Expert systems with applications Ročník 238; s. 121812
Hlavní autoři: Xu, Shengbing, Hao, Zhifeng, Zhu, Yuanhao, Wang, Zhenyou, Xiao, Yunhao, Liu, Bo
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 15.03.2024
Témata:
Constraints
Fuzzy clustering
Missing value
Prior membership degree
Semi-supervised clustering
Fuzzy clustering
Constraints
Missing value
Semi-supervised clustering
Prior membership degree
ISSN:0957-4174, 1873-6793
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:Existing pre-processing methods for the prior membership degree matrix suffer from the following issues: (1) The labeling constraints for prior membership degree matrix have an effect on the expert’s judgment on the prior membership degree, which easily causes the distortion problem of the prior membership degree labeling information; (2) There exists the problem of inconsistency between the filling information and the labeling information in the prior membership degree matrix to be filled in the missing values with zeros. To address these problems, we propose an unconstrained labeling idea for the prior membership degree matrix and the corresponding pre-processing method for the missing values by introducing the statistical characteristics of extreme value distribution and simultaneously apply it to the semi-supervised fuzzy clustering algorithm. More specifically, we focus on learning an expert preference value from the prior membership degree matrix and filling in the missing values with the expert preference value. Thus, we propose an unconstrained pre-processing method for the prior membership degree matrix by filling in missing values with an expert preference to keep the filling information consistent with the labeling information in the prior membership degree matrix as much as possible. In addition, we design a semi-supervised fuzzy clustering algorithm based on an unconstrained prior membership degree matrix with expert preference (SFCM-EP) by introducing the K-L divergence to improve the applicability, utility and running performance of semi-supervised fuzzy clustering algorithm. Our experimental results on the simulation dataset and the UCI datasets show the feasibility and effectiveness of the proposed pre-processing method of the prior membership degree matrix with encouraging results.
ISSN:0957-4174
1873-6793
DOI:10.1016/j.eswa.2023.121812