Hybrid neural network framework for detection of cyber attacks at smart infrastructures

Last decade, the speed of wireless data transmission has increased dozens of times, while in the near future a widespread introduction of 5G/5G+ technology is expected which will significantly improve the stability and reliability of inter-device communications in smart cyber infrastructures: Internet of Things (IoT/IIoT), e-hospitals, Factory-of-the-Future (FoF). The presence of high-speed and reliable wireless communications leads to a rapid jump in the creation of modern machine-to-machine smart infrastructures. However, such an environment provides new opportunities for intruders to commit new cyber attacks, e.g. Black hole, Gray hole, massive DDoS, which can lead to serious consequences in dynamic network routing. Researchers thus face the challenge of creating new methods for countering cyber attacks of such kinds. One of the approaches is to apply artificial neural network (ANN). We suggest to utilize the modern ANNs corresponding security tasks in dynamic infrastructures. The paper identifies their major advantages and evaluates the possibility of their application for solving the issue of accurate attacks detection at machine-to-machine (m2m) adhoc self-organizing networks. An assessment is made of the effectiveness of the activation functions and optimization methods to improve accuracy in detecting the network routing attacks. A complex neural framework protecting the m2m-network against attacks on dynamic routing has been discussed, and accuracy of the developed hybrid is estimated. The ensemble of modern ANNs demonstrates 96% accuracy in detecting all types of routing attacks. This neural network complex can be applied to protect smart infrastructures against linkage collapses and network disunity.