Design Graph-Structured Dataset and Feature Selection for Cyber Threat Detection

In the digital era, technology rapidly evolves, becoming integrated into devices such as mobile phones, tablets, and laptops. Cyberattacks continuously evolve and adapt to exploit potential vulnerabilities in emerging technologies. The types of cyberattacks have become increasingly complex and diverse, posing significant threats to the security of digital assets and personal information. Therefore, machine learning techniques for analyzing threats represented as graphs have emerged to respond to complex attacks. For this reason, it remains challenging to develop and design datasets with a graph-based structure. The majority of open-source datasets related to cybersecurity come from capturing network traffic, which may not fully encompass comprehensive cyber threat detection. The methodology involves a process of feature extraction from operating system logs. This is essential for facilitating processing in the form of graph-based machine learning. The results will consist of two types of datasets: Edge Dataset collects data in the form of edges, which capture the connectivity relationships between device networks. Node Dataset collects data in the form of nodes, containing features and labels, further subdivided into Windows and Ubuntu operating systems. The dataset will be used for cyber threat detection and made available to other researchers interested in machine learning for further development related to graphs.