Targeted ranking-based clustering using AHP K-means
K-Means can group similar objects features into specified number (K) of cluster centers region. Similarity is measured based on their closest distance of multiple features coordinate location. However, such distance measurement can be doubtful in satisfying certain clustering application as it does...
Saved in:
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
International Center for Scientific Research and Studies
2015
|
Subjects: | |
Online Access: | http://eprints.unisza.edu.my/6922/1/FH02-FIK-15-04680.jpg http://eprints.unisza.edu.my/6922/ |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Summary: | K-Means can group similar objects features into specified number (K) of cluster centers region. Similarity is measured based on their closest distance of multiple features coordinate location. However, such distance measurement can be doubtful in satisfying certain clustering application as it does not distinguish the meaning of object features representation. Ordinal feature for example may denote to certain ranking objects rather than just number representation. Thus, clustering result should also consider the existing rank label on these objects instead of distance measurement. New AHP K-Means technique is proposed to preserve rank order for each object in the clustering result. It transforms weighted multi-features objects by aggregating them as a single ranking objects using pair-wise comparing among the objects. These ranking objects are then processed by K-Means based on cluster centers that initially setup on fair distributed ranking scale. Based on experiment using weighted course marks of 92 students, the proposed technique shows that ranking-based clustering using AHP can give accurate ranked clustering result compared to normal weighted K-Means. |
---|