Efficient Anomaly Detection Methodology for Power Saving in Massive IoT Architecture

Energy saving is the paramount factor in the evolving Internet of Things (IoT) due to limited battery energy in the devices. An IoT device in anomaly condition will lead to transmission failure and power drain. It is imperative to detect the anomaly, whose occurrence is random in nature over time. The randomness in failure needs a statistical model of an IoT device to predict and stop the occurrence of anomaly. We propose a novel approach of modeling IoT devices as a finite state automaton, which is irreducible, a priori, has well determined emissions (refers to received signal strength) and finite hidden state space. We have designed a Hidden Markov Model (HMM) based approach to efficiently predict anomaly using which we orchestrate the time interval between successive transmissions from an IoT device. Experimental results reveal that our approach can determine anomaly of IoT device with accuracy as high as 98%. The higher anomaly detection rate results in saving around 14% of IoT device battery power by avoiding redundant transmissions.