Distributed probability density based multi-objective routing for Opp-IoT networks enabled by machine learning

Abstract

 Opportunistic IoT networks operate in an intermittent, mobile communication topology, employing peer-to-peer transmission hops on a store-carry-forward basis. Such a network suffers from intermittent connectivity, lack of end-to-end route definition, resource constraints and uncertainties arising from a dynamic topology, given the mobility of participating nodes. Machine learning is an instrumental tool for learning and many histories-based machine learning paradigms like MLPROPH, KNNR and GMMR have been proposed for digital transformations in the field with varying degrees of success. This paper explores the dynamic topology with a plethora of characteristics guiding the node interactions, and consequently, the routing decisions. Further, the study ascertains the need for better representation of the versatility of node characteristics that guide their behavior. The proposed scheme Opportunistic Fuzzy Clustering Routing (OFCR) protocol employs a three-tiered intelligent fuzzy clustering-based paradigm that allows representation of multiple properties of a single entity and the degree of association of the entity with each property group that it is represented by. Such quantification of the extent of association allows OFCR a proper representation of multiple node characteristics, allowing a better judgement for message routing decisions based on these characteristics. OFCR performed 33.77%, 6.07%, 3.69%, 6.88% and 78.14% better than KNNR, GMMR, CAML, MLPRoPH and HBPR respectively across Message Delivery probability. OFCR, not only shows improved performance from the compared protocols but also shows relatively more consistency across the change in simulation time, message TTL and message generation interval across performance metrics.