Transactions on Machine Learning and Data Mining (Print ISSN: 1865-6781) (ONLINE-ISSN: 2509-9337) (ISBN: 978-3-942952-86-6 )
Volume 14 - Number 1 - July 2021 - Pages 31-44
Comparison of Methods for Computing Similarity Based on Clusters - Utilizing Different Membership Functions
Arthur Yosef 1, Eli Shnaider2, and Moti Schneider3
1Tel Aviv-Yaffo Academic College, Israel,
2Independent Researcher, Israel
3Netanya Academic College, Israel
Abstract
In this paper, we compare various methods for computing similarity between numerical vectors based on their division into clusters. The advantage of utilizing clusters is apparent mostly in the cases where the data are very unreliable and distorted, so that the cluster represents approximate value of its elements in a very broad term. Measuring similarity between numerical vectors following their division into clusters, provides additional method for similarity measurement, which might be a preferable method when lack of confidence in the measurements of individual data elements is high. In addition, we compare the influence of applying the various types of membership functions on the results of similarity measurements.
Keywords: Data Mining, Soft Computing, Fuzzy Logic, Similarity Measure, Clusters.
Download Paper (1106 KB)