Bipolar fully recurrent deep structured neural learning based attack detection for securing industrial sensor networks

Attack detection is a significant problem to be resolved to attain security in industrial sensor network. Few research works have been designed for performing attack discovery process with the help of different classification algorithms. However, classification performance was lowered to accurately find attack nodes and thereby getting higher security level (SL). In order to overcome such limitations, bipolar fully recurrent deep structured neural learning based attack detection (BFRDSNL-AD) technique is proposed in this article. The BFRDSNL-AD technique comprises of three main layers, namely, input, hidden, and output layer to accurately carry-out attack detection process. The input layer in BFRDSNL-AD Technique takes a number of sensor nodes as input. The designed BFRDSNL-AD technique utilized numbers of hidden layers in order to deeply examine each sensor node in industrial sensor network. The result of the hidden layer is fed-back into the network along with the inputs in order to discover the temporal dynamic behavior of sensor nodes. In BFRDSNL-AD technique, output layer applies bipolar activation function to determine the association between identified node features to give attack detection result. If the output layer result is 1, then sensor node is considered as normal. Otherwise, sensor node is considered as malicious attack in industrial sensor network. Thus, the BFRDSNL-AD technique increases the attack detection performance with higher accuracy and minimal time. From that, BFRDSNL-AD technique gets enhanced SL in industrial sensor network. The simulation of BFRDSNL-AD technique is conducted using metrics such as security, data delivery rate, data loss rate, and delay with respect to a different number of sensor nodes and data packets.