Neural Network Based Forecasting Technique for Wireless Sensor Networks

The diversified and huge applicability of sensor networks has attracted the researchers in this field. The nodes in the sensor networks are distinguished by the scarce resources; hence energy conservation approaches are of great significance. The node scheduling approaches aims to schedule the nodes in a number of set covers, which can be activated periodically to monitor the given points of interest with the desired confidence level along with the objective of maximizing coverage and network lifetime. The determination of set covers is considered as a NP hard problem and is dependent on different network parameters such as node contribution, trust values and coverage probability.. The main motivation of the proposed approach is to reduce this complexity by employing the prediction technique based on learning through neural network. The paper presents the neural network based prediction model to determine the activation status of the nodes in the set cover. In this scheme, the node has to monitor the neighboring node parameters at regular intervals, which incurs a huge number of communications and overhead. The data prediction technique can reduce this overhead by autonomously determining the node activation status. The paper proposes a neural network-based prediction technique for sensor networks in combination with the node scheduling strategy. The different node parameters are provided as input to train the network for prediction of node status. The performance of the different prediction models have been evaluated in terms of precision, recall, f1 score and accuracy for the training and test datasets. The binary cross entropy-based loss function is analyzed in training the neural networks. The accuracy of the model is evaluated for the validation split size as 20%. The simulation results show that the accuracy in the prediction of the node status is maximum for the NAdam based optimizer i.e. 87% and 76% for the training and the testing dataset respectively.