Clustering Algorithms for Enhanced Trustworthiness on High-Performance Edge-Computing Devices

Abstract

Trustworthiness is a critical concern in edge-computing environments as edge devices often operate in challenging conditions and are prone to failures or external attacks. Despite significant progress, many solutions remain unexplored. An effective approach to this problem is the use of clustering algorithms, which are powerful machine-learning tools that can discover correlations within vast amounts of data. In the context of edge computing, clustering algorithms have become increasingly relevant as they can be employed to improve trustworthiness by classifying edge devices based on their behaviors or detecting attack patterns from insecure domains. In this context, we develop a new hybrid clustering algorithm for computing devices that is suitable for edge computing model-based infrastructures and that can categorize nodes based on their trustworthiness. This algorithm is thoroughly assessed and compared to two computing systems equipped with high-end GPU devices with respect to performance and energy consumption. The evaluation results highlight the feasibility of designing intelligent sensor networks to make decisions at the data-collection points, thereby, enhancing the trustworthiness and preventing attacks from unauthorized sources.