Deep Q learning based secure routing approach for OppIoT networks

Abstract

Opportunistic IoT (OppIoT) networks are a branch of IoT where the human and machines collaborate to form a network for sharing data. The broad spectrum of devices and ad-hoc nature of connections, further alleviate the problem of network and data security. Traditional approaches like trust based approaches or cryptographic approaches fail to preemptively secure these networks. Machine learning (ML) approaches, mainly deep reinforcement learning (DRL) methods can prove to be very effective in ensuring the security of the network as they are profoundly capable of solving complex and dynamic problems. Deep Q-learning (DQL) incorporates deep neural network in the Q learning process for dealing with high-dimensional data. This paper proposes a routing approach for OppIoT, DQNSec, based on DQL for securing the network against attacks viz. sinkhole, hello flood and distributed denial of service attack. The actor–critic approach of DQL is utilized and OppIoT is modeled as a Markov decision process (MDP). Extensive simulations prove the efficiency of DQNSec in comparison to other ML based routing protocols, viz. RFCSec, RLProph, CAML and MLProph.