Federated Multi-View K-Means Clustering.

Area of Science:

• * Computer Science
• * Data Science
• * Machine Learning

Background:

• * The Internet of Things (IoT) generates vast amounts of Big Data, often non-independently and identically distributed (non-IID).
• * Existing clustering algorithms struggle with data privacy and the non-IID nature of Big Data in distributed environments.

Purpose of the Study:

• * To develop a privacy-preserving multi-view k-means (MVKM) clustering algorithm for federated learning environments.
• * To enhance MVKM by integrating exponential distance for weighted Euclidean distance calculations, leveraging federated learning advancements.

Main Methods:

• * Introduction of a novel federated MVKM (Fed-MVKM) algorithm operating on local client data principles.
• * Integration of exponential distance transformation to adapt weighted Euclidean distance for MVKM.
• * Implementation using synthetic and six real multi-view datasets, with data splitting via Federated Peter-Clark (Huang et al., 2023).

Main Results:

• * Fed-MVKM demonstrates suitability for clustering large datasets in a federated setting.
• * Shared models based on local cluster centers with data-driven approaches yield improved clustering performance.
• * The algorithm successfully generates a satisfying final pattern for multi-view data, outperforming non-federated MVKM.

Conclusions:

• * The proposed Fed-MVKM algorithm effectively addresses privacy concerns in Big Data clustering within federated environments.
• * Fed-MVKM offers a novel approach to multi-view clustering, enhancing performance through federated learning and data-driven local centers.
• * This method provides a scalable and efficient solution for analyzing large, distributed, and non-IID multi-view datasets.