A survey on Iot security: Application Areas, Security Threats, and Solution Architectures
VII. IOT SECURITY USING MACHINE LEARNING
tải về 6.16 Mb. Chế độ xem pdf
|
thong-tin-so bao-cao-qpsk-1 - [cuuduongthancong.com]
- Điều hướng trang này:
- A. SOLUTIONS PROVIDED BY ML TO OVERCOME SECURITY THREATS
| VII. IOT SECURITY USING MACHINE LEARNING
The area of machine learning (ML) has attracted significant interest over recent years. Many domains are using ML for their development, and it is being used for IoT security as well. ML appears to be a promising solution to protect IoT devices against cyber attacks by providing a different approach for defending against attacks as compared to other traditional methods. A. SOLUTIONS PROVIDED BY ML TO OVERCOME SECURITY THREATS In regard to the attacks discussed in Section III, the solutions provided by ML to overcome these security threats are dis- cussed below. 1. DoS Attack: DoS attacks on IoT devices or origi- nating from IoT devices are a serious concern. One approach to prevent such attacks is to use a Multi- Layer Perceptron (MLP) based protocol that secures networks against DoS attacks [178]. The authors of [179] have proposed a particle swarm optimization and back propagation algorithm to train a MLP that helps in enhancing the security of wireless networks. ML techniques help in increasing the deduction accuracy and securing IoT devices vulnerable to DoS attacks. 2. Eavesdropping: Attackers may eavesdrop on mes- sages during data transmission. To provide protection from such attacks, ML techniques such as Q-learning based offloading strategy [180] or non-parametric Bayesian techniques [181] can be used. Schemes such as Q-learning and Dyna-Q are ML techniques that may also be used to protect devices from eavesdropping. Evaluation of these schemes via experiments and re- inforcement learning is presented in [182]. 3. Spoofing: Attacks from spoofers can be avoided by using Q-learning [182], Dyna-Q [182], Support Vector Machines (SVM) [183], Deep Neural Network (DNN) model [184], incremental aggregated gradient (IAG) [46], and distributed FrankWolfe (dFW) [185] tech- niques. These techniques not only increase the detec- tion accuracy and classification accuracy but also help in reducing the average error rate and false alarm rate. 4. Privacy Leakage: Collection of personal information such as health data, location, or photos puts the user’s privacy at stake. Privacy-preserving scientic computa- tions (PPSC) [186] should be employed for preventing privacy leakage. A commodity integrity detection algo- rithm (CIDA) which is based on the Chinese remainder theorem (CRT) is another technique that has been proposed to develop IoT application trust [187]. 5. Digital Fingerprinting: Digital fingerprinting is one of the upcoming and promising solutions to secure IoT systems and to help the end users gain sufficient trust in the applications. Fingerprints are being widely used to unlock smartphones, approve payments, unlock the car and home doors, etc. Due to its low cost, reliability, acceptability and high-security level, digital fingerprinting is emerging as a dominant bio-metric identification method [188]. Apart from the benefits of digital fingerprinting, there are various challenges to efficiently use this technique in IoT, such as fingerprint classification, image enhancement, feature matching, etc. Various machine learning based algorithms have been developed to provide some non-traditional so- lutions to overcome these challenges [189], some of 16 VOLUME x, 2019 This work is licensed under a Creative Commons Attribution 3.0 License. For more information, see http://creativecommons.org/licenses/by/3.0/. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2019.2924045, IEEE Access Vikas Hassija et al.: A Survey on IoT Security: Application Areas, Security Threats, and Solution Architectures which are discussed below. • Support Vector Machine: SVM is a training al- gorithm for non-linear and linear classifications, principal component analysis, text categorization, speaker identification, and regression. It maxi- mizes the gap between the decision boundary and training patterns. Authors of [190] have discussed the use of SVM in digital fingerprinting in detail. They have also compared it with other traditional models. A feature vector is built based on pixel values of the fingerprint, and it is used to train the SVM. Various patterns behind the fingerprint are analyzed, and then the matching of a fingerprint is done based on patterns identified. • Artificial Neural Networks: ANN is one of the most commonly used algorithms in the machine learning. It offers many advantages like fault tolerance, adaptive learning, and generalization. In [191] a framework has been proposed for using ANN to identify fingerprints digitally. The digital values of various features in the fingerprint like minutiae, ridge ending, and bifurcation is applied as the input to the neural network for training pur- pose using back propagation algorithm of ANN. The verification of the fingerprint is done based on the previous experiential values stored in the database. The fundamental need in IoT is to secure all the systems and devices that are connected to the network. The role of ML is to use and train algorithms to detect anomalies in IoT devices or to detect any unwanted activity taking place in IoT system to prevent data loss or other issues. Therefore, ML provides a promising platform to overcome the difficulties faced in securing IoT devices. Further contributions in this field are required to maintain the growth of IoT. tải về 6.16 Mb. Chia sẻ với bạn bè của bạn:
|