Showing 1–1 of 1 results for author: Son, S B

Entropy-Aware Similarity for Balanced Clustering: A Case Study with Melanoma Detection

Authors: Seok Bin Son, Soohyun Park, Joongheon Kim

Abstract: Clustering data is an unsupervised learning approach that aims to divide a set of data points into multiple groups. It is a crucial yet demanding subject in machine learning and data mining. Its successful applications span various fields. However, conventional clustering techniques necessitate the consideration of balance significance in specific applications. Therefore, this paper addresses the… ▽ More Clustering data is an unsupervised learning approach that aims to divide a set of data points into multiple groups. It is a crucial yet demanding subject in machine learning and data mining. Its successful applications span various fields. However, conventional clustering techniques necessitate the consideration of balance significance in specific applications. Therefore, this paper addresses the challenge of imbalanced clustering problems and presents a new method for balanced clustering by utilizing entropy-aware similarity, which can be defined as the degree of balances. We have coined the term, entropy-aware similarity for balanced clustering (EASB), which maximizes balance during clustering by complementary clustering of unbalanced data and incorporating entropy in a novel similarity formula that accounts for both angular differences and distances. The effectiveness of the proposed approach is evaluated on actual melanoma medial data, specifically the International Skin Imaging Collaboration (ISIC) 2019 and 2020 challenge datasets, to demonstrate how it can successfully cluster the data while preserving balance. Lastly, we can confirm that the proposed method exhibited outstanding performance in detecting melanoma, comparing to classical methods. △ Less

Submitted 11 May, 2023; originally announced May 2023.