A Convergence Time Predictive Model using Machine Learning for LLN
Authors: Garg, S., Mehrotra, D., Pandey, S. and Pandey, H.M.
Journal: 2021 IEEE 8th Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering, UPCON 2021
ISBN: 9781665409629
DOI: 10.1109/UPCON52273.2021.9667637
Abstract:The need to interface Low power and Lossy Network (LLN) to the web acquired notoriety with the rise of Internet of Things (IoT). Accordingly, IETF ROLL working group proposed a de-facto IPv6 routing protocol called RPL. RPL provisions 6LoWPAN (IPv6 over Low power and Wireless Personal Area Network) and has been the profound interest among researchers, primarily because of its flexibility to cope with the topology changes and its ability to auto-configure, detect and avoids loops. Since, motes that are deployed in IoT network are battery driven and lossy in nature, network execution is strongly influenced. Consequently, if the network convergence for scalable network can be foreseen, it can be utilized to upgrade the network performance. The idea behind this article is to propose a predictive model that gauges Convergence Time (CT) by performing feature selection by utilizing Machine Learning (ML) strategy for RPL and IoTorii. IoTorii is another such convention proposed in recent literature that scales network better. RPL and IoTorii protocols are simulated on Contiki OS/Cooja simulator using Sky motes. Further, RPL execution precision is tried for Storing and Non-Storing modes both. Similarly, IoTorii performance accuracy is tested for both of its proposed variations: nHLMAC1 and nHLMAC3 addresses. Additionally, the network parameters obtained from the simulation are used for feature selection in predictive modelling. The experiment shows that the prediction model gives the best forecast with 93.619%, 96.962%, 93.112% and 92.635% accuracy for both the protocols with different modes and addresses respectively.
Source: Scopus