TOWARDS INTRUSION DETECTION IN IOT USING FEW-SHOT LEARNING

Abstract

The Internet of Things (IoT) is an emerging technology that covers various domains and has become an essential part of the upcoming technological revolution. IoT applications include healthcare, smart-cities, smart-cars, industries, quality of life, and several other fields. IoT typically consists of lightweight sensor devices that facilitate procedures such as automation, real-time trackable data collection, and data-driven decisions. However, securing IoT networks is an accessible research area for several reasons. The main security challenges are limited resources that are incapable of dealing with complex and advanced security tools; and lack of required data for training the security systems like Intrusion detection systems as a result of their heterogeneous nature. This research proposed a Few-shot learning IoT intrusion detection system model based on a Siamese network to overcome the above limitation. The model aims to classify and distinguish normal and attacked traffic. The experiment utilized an IoT dataset in different scenarios to analyze and validate the behavior with three categories with different numbers of data in each. The performance result achieves more than 99% accuracy and shows an efficient detection ability using only less than 1% of the dataset.