Abstract
Federated learning has emerged as a powerful paradigm for collaborative machine learning across distributed agents without centralized data aggregation. While successful in supervised tasks, its extension to unsupervised learning, such as clustering, faces challenges due to the absence of labels, data heterogeneity, and the need for algorithmic flexibility. These challenges are especially critical in autonomic systems, where agents such as drones, robots, and IoT devices operate with limited resources and dynamic, unpredictable data that emerge at runtime. To address these challenges, we propose, apply, and assess FedMAC (Federated Multi-Algorithm Clustering), a decentralized clustering framework tailored for autonomic systems. FedMAC allows clients to perform local clustering using their preferred centroid-based algorithms (e.g., k-means, Mean Shift) and variable cluster counts. Clients share only cluster centroids and metadata, preserving privacy, while a central server aggregates this information through a similarity-based merging strategy to build a global clustering model. We evaluate FedMAC on a synthetic dataset simulating evolving non-IID data distributions. FedMAC achieves up to 2 5. 1 % higher silhouette scores under constrained computation (5 local iterations) and shows a 21.2 % relative gain as the number of clients increases from 4 to 16. These results demonstrate the framework's scalability, privacy preservation, and ability to reconstruct coherent global patterns, making it well-suited for collaborative mapping, distributed perception, and swarm coordination in federated autonomic systems.