Adaptive Novelty Detection over Contextual Data Streams at the Edge using One-class Classification

Abstract

Online novelty detection is an emerging task in Edge Computing trying to identify novel concepts in contextual data streams which should be incorporated into predictive analytics and inferential models locally executed on edge computing nodes. We introduce an unsupervised adaptive mechanism for online novelty detection over multi-variate data streams at the network edge based on the One-class Support Vector Machine; an instance of One-class Classification paradigm. Due to the proposed adjustable periodic model retraining, our mechanism timely and effectively recognises novelties and resource-efficiently adapts to data streams. Our experimental evaluation and comparative assessment showcase the effectiveness and efficiency of the proposed mechanism over real data-streams in identifying novelty conditioned on the necessary model retraining epochs.