Related Experiment Video
Updated: Aug 6, 2026
06:08
Automatic Identification of Dendritic Branches and their Orientation
Published on: September 17, 2021
Federated parameter-free DBSCAN clustering and its application in image recognition
Fang Cheng1, Zilong Deng1,2, Mustafa Muwafak Alobaedy2
1College of Information Technology, Anqing Vocational and Technical College, Anqing, China.
Plos One
|August 4, 2026
Summary
We introduce two novel federated DBSCAN algorithms, FDBSCAN and FDBSCAN++, for privacy-preserving clustering of distributed data. These methods enhance clustering accuracy in edge computing environments.
Area of Science:
- Computer Science
- Artificial Intelligence
- Data Mining
Background:
- Clustering distributed data with privacy is a significant challenge for traditional DBSCAN.
- Edge computing environments require efficient and secure data processing methods.
Purpose of the Study:
- To propose secure, federated, parameter-free DBSCAN clustering methods for edge computing.
- To address privacy concerns in distributed data clustering.
Main Methods:
- Combining federated clustering with DBSCAN (Density-Based Algorithm for Discovering Clusters in Spatial Databases with Noise).
- Applying differential privacy to client data and using adaptive DBSCAN to identify core points.
- Clients send core points to a server for aggregation into global cluster centers.
Main Results:
- FDBSCAN and FDBSCAN++ demonstrate superior clustering accuracy compared to traditional and SOTA algorithms across eight real datasets, including MNIST.
- The algorithms achieve satisfactory results in image clustering and segmentation tasks.
Conclusions:
- FDBSCAN and FDBSCAN++ offer effective solutions for privacy-preserving distributed data clustering in edge environments.
- The proposed methods advance the application of DBSCAN in complex, distributed settings.