Behavioral analysis in Cybersecurity using Machine Learning. A study based on graph representation, class imbalance and temporal dissection

Nowadays, cybersecurity plays a key role in everyday life. Each day more and more devices are interconnected sharing information, communications, and connections. Paradigms like IoT, Industry 4.0, Smart Factory, 5G, have increased the velocity of such growth. According to [LPS21], there are about 20 billion devices connected to the global net, like wearables, medical devices,
automotive control units, smartphones, televisions, fridges, and so on. In this scenario, the attack surface [ROC+20] has increased enhancing the number of possible threats and vulnerabilities that can be exploited by a cyber-attack. A cyber-attack can be seen as an attack on the Confidentiality, Availability, and Integrity (CIA triad) of an information system, network, software, etc. [SC14].
The consequences of a cyber-attack are not only limited to digital leaks or information losses, but can have a very huge impact in economical, ethical, digital, and societal terms for the attacked company. For this reason, data protection and the management/monitoring of information systems has become a primary task for many companies [MPDH19], as well as for the European Community.