Denial of Service (DoS) Defences against Adversarial Attacks in IoT Smart Home Networks using Machine Learning Methods

Abstract

The availability of information and its integrity and confidentiality are important factors in information and communication of the system security. The DDoS attack generally means Distributed denial of services generates many enormous packets to slow and down the Services for actual users who use services. The study examines the impact of a considerable rise in the number of connected devices in the IoT concept on the quantity and volume of DDoS attacks. Thanks to machine learning-based technology, intrusion Detection Systems (IDS) can be versatile and efficient. However, the advancement of machine learning systems, alongside the application of the uses for Adversarial Machine Learning, has also introduced a new potential attack vector; machine learning models which support the uses of the IDS’ decisions may be subject to cyberattacks known as Adversarial Machine Learning (AML). AML is widely applicable to manipulating data and network traffic that transverse networked devices in the IoT setting. However, harmful network packets are frequently misclassified as benign perturbations in the machine learning classifier’s decision bounds. Because of this, machine learning-based detectors such as malware scanners skip those flaws and reduce the risk of delaying detection and spreading malicious code, and incurring issues such as personal information leaking, damaged hardware, and financial loss. Furthermore, this research investigates which DoS attack techniques should be implemented and how adversary samples should be constructed to strengthen the robustness of supervised models utilizing adversarial training. The system obtained 99.98% accuracy with XGBoost and 99.96% accuracy achieved with the decision tree and AdaBoost.