SecureDeepNet‐IoT: A deep learning application for invasion detection in industrial Internet of Things sensing systems

Abstract

Deep learning (DL) is a special field of artificial intelligence that has increased its use in various fields and has proved its effectiveness in classification. The feasibility of using many hidden layers and many neurons for each layer in the DL architectures enables a detailed analyzing capability for classification and segmentation issues. Advancing the learning performance and generalization capability of the models for big data is one of the major advantages of DL that makes it a mandatory requirement for especially the Internet of Things (IoT) technology. Whereas tethering the sensing systems to the Internet enables remote control capabilities, communication, and coordination among the systems raises security vulnerability for many IoT devices. This study utilized DL algorithms, including deep belief networks and deep autoencoders, to perform an invasion detection application in IoT devices. The UNSW‐NB15 database, collected from a realistic network to evaluate cybersecurity applications, was utilized for training the DL models. The study aims to design a DL architecture for the implementation of secured IoT and to assess the performance and robustness of DL algorithms for detecting IoT‐based intrusions. The main contribution is SecureDeepNet‐IoT, which is an adaptive multikernel cybersecurity platform for sensing inputs. The SecureDeepNet‐IoT has achieved classification accuracy up to 95.05% and 94.39% for deep belief networks and deep neural networks, respectively. Deep autoencoder with extreme learning machine kernel has detected the IoT‐based invasion with accuracy rates of 97.86% and 98.15% for binary and multiinvasion classification, respectively. In addition to high generalization capability, the deep autoencoder kernel reached rapid enough speed in detecting abnormal sensing values in IoT.

An energy‐aware and lightweight routing algorithm using clustering and mobile sinks in wireless sensor networks

“…Simulation results show the better performance of ERCMS against DFCR, 58 MDFCR and TEO‐MCRP 35 in terms of (a) consumed energy, (b) delivery ratio, (c) delay, (d) routing overhead, (e) throughput, and (f) the first and the last node's death times performance metrics. It is worth to mention that, by decreasing energy consumption and increasing network lifetime more advanced security schemes (like References 73,74) can be utilized. In the future studies we will explore a way to learn coefficient $$$ \theta $$$ (sectoring angle) and $$$ {B}_t $$$(number of mobile sinks considered by the network designer to support the received help requests in each time interval t ) of our algorithm.…”

An optimized hybrid deep neural network architecture for intrusion detection in real‐time IoT networks

“…Cyber and physical attacks are two kinds of attacks involved in smart cities 5‐10 . Sleep Denial Attack, Side Channel Attack, Permanent Denial of Service, Fake Node Injection, Radio Frequency Jamming, and Malicious Code Injection are examples of these assaults 11 .…”

“…4 Cyber and physical attacks are two kinds of attacks involved in smart cities. [5][6][7][8][9][10] Sleep Denial Attack, Side Channel Attack, Permanent Denial of Service, Fake Node Injection, Radio Frequency Jamming, and Malicious Code Injection are examples of these assaults. 11 In a cyber-attack, the attacker attempts to inject malicious or malware software into network components in order to gain unauthorized access to them.…”

Android Malware Application Detection using Multi-layer Perceptron

“…Spyware can be used to obtain GPS locations, correspondence, and bank information of individuals, as well as to obtain consumer habits for customized advertisements [2]. Moreover, advanced artificial intelligence algorithms are adapted to sensing systems in IIoT networks for detecting malicious applications and various cyber-attack from network traffic [17].…”