ReShape: A Universal Data Processing Approach for Clustering Inhomogeneous and Weakly Separated Manifold Data
- 20 May 2024
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Industrial Informatics
- Vol. 20 (9) , 10901-10911
Abstract
Classical clustering methods such as k-means, spectral clustering, DBSCAN, and DPC often operate under specific assumptions about data characteristics. However, the increasing diversity and complex characteristic of data, coupled with limited prior knowledge, hinder the direct application of these clustering algorithms. Existing data processing approaches, such as ReScale and CDF-TS, are capable of homogenizing cluster density, but tailor to density-based algorithms that does not perform well on weakly separated or manifold clusters. To overcome these issues, we propose a novel data processing method (ReShape) facilitating the direct application of well-known clustering algorithms without modifications. The ReShape integrates three key components: a homogenization factor for uniforming varying cluster densities, a separation factor for enhancing cluster separations, and a joint distance considering both global and local consistency to measure distances in manifold datasets. The proposed method is universally applicable, and can match implicit assumptions of existing clustering algorithms. Extensive experiments and a case study for fault diagnosis involving k-means, spectral clustering, DBSCAN, and DPC demonstrates the efficacy of the ReShape approach.Keywords
Funding Information
- National Natural Science Foundation of China (72271120, 72001107)
- China Postdoctoral Science Foundation (2020T130297, 2019M660119)
This publication has 30 references indexed in Scilit:
- Density peak clustering algorithms: A review on the decade 2014–2023Expert Systems with Applications, 2023
- Density Peak Clustering with connectivity estimationKnowledge-Based Systems, 2022
- CDF Transform-and-Shift: An effective way to deal with datasets of inhomogeneous cluster densitiesPattern Recognition, 2021
- Dataset-Transformation: improving clustering by enhancing the structure with DipScaling and DipTransformationKnowledge and Information Systems, 2019
- Fuzzy clustering with nonlinearly transformed dataApplied Soft Computing, 2017
- Density-ratio based clustering for discovering clusters with varying densitiesPattern Recognition, 2016
- Natural neighbor: A self-adaptive neighborhood method without parameter KPattern Recognition Letters, 2016
- Spectral clustering with density sensitive similarity functionKnowledge-Based Systems, 2011
- Data clustering: 50 years beyond K-meansPattern Recognition Letters, 2010
- Robust path-based spectral clusteringPattern Recognition, 2008