Work

Inferring the behaviour and security of networked devices via communication analysis

Year: 2019
Type: dissertation
Abstract: In the last years the number of devices connected to the Internet has been increasing exponentially and it has reached huge numbers, just days ago the RIPE Network Coordination Center announced it run out of IPv4 addresses. \nAs a consequence, it's extremely complex to control what is happening on the network and what, or whom, a device is communicating with. We are losing control over many of our devices. It's becoming everyday more difficult to know what the device is doing and what is it sharing on the Internet. \nLuckily, in the last years the attention to security, privacy and awareness is ever increasing: the users pay more attention to what their devices are doing and care about it. \nWhile big software (like the operative systems and most mobile devices) make available an increasing number of tools to monitor the device traffic, some devices cannot be inspected or those same tools may be forged in a way that makes it impossible to detect some specific malicious behaviours. \nUntil now every successful method to detect or filter the behaviour of the devices has been trying to only detect anomalies or manually lock some specific behaviours; this means nobody ever tried to understand at a discrete \nlevel of accuracy what the device does relying only on the analysis of the metadata of the traffic intercepted. \nThat’s why in this thesis I defined a model able to detect the behaviour which is occurring on the device by the mere observation of its network traffic. In doing so, I first defined a model that takes the raw low-level information regarding the communications occurring, process them and return information about the high-level operations occurring on the device. I then built a demo that uses that model and demonstrates its feasibility. The data used are low-level information on the traffic without ever inspecting its payload. \nTo improve the efficiency of the algorithm, I also use a form of Recurrent Neural Networks to finally infer the device behaviour.
Author: Pierfederici, Eugenio
Language: English
Open Access status: green
OpenAlex ID: W7066471593
Cites: 18
Cited by:
Related to: 10