Behavioral Modeling Of Sensor Nodes Using Meta-data

Abstract

Most of the Wireless Sensor Network(WSN) systems are deployed under extreme conditions. Thus, it is important to maximize their lifespan by optimal use of network's resources. Battery lifespan of a node is a crucial resource that need to be used carefully. Energy aware routing and schedule sensing approaches have been introduced for the careful use of battery. Optimization of battery usage can be achieved through predicting the future behavior of its energy consumption. This will lead the users and applications to take early decisions, thus minimizing network downtime. Hence, we explore the possibility of using meta-data, which are unique to each node, to represent and predict node behavior using machine learning models. In this research, we use node voltage level as an indicator of the energy used, as voltage is proportional to available energy. Node energy consumption is modeled and predicted by ARIMA models using these voltage readings. We also classify node with respect to its current and future usage, thus allowing user to take early decisions maximizing network throughput (lifetime). After evaluating our results against a created base set of behavioral classi cations, it showed an 80% accuracy rate when we classify nodes to three classes as high use, medium use, and low use. Our forecasting method produces an average Mean Absolute Percentage Error(MAPE) rate of 0.38%, and a Root Mean Squared Error(RMSE) of 0.01, for the generalized result. By being able to predict a node's energy consumption behavior at a higher accuracy, energy aware applications and protocols can take optimization decisions beforehand to increase the network lifetime.