Sec4ML: anonimização de dados de incidentes de segurança da informação para tarefas de aprendizado de máquina

Abstract

Despite the exponential growth of the World Wide Web since its creation, there are still few available datasets of cybersecurity incidents to be reused due to several issues, such as privacy-preserving concerns and data publication format standardization. As a result, the domain incidents analysis are precarious impacting on the Intrusion Detection Systems (IDS) development. The LOD (Linked Open Data) practices, which allows the sharing of data on the Web as a large and interconnected data graph, together with the FAIR (Findable, Accessible, Interoperable, and Reusable) principles, which guides the publication of data for reuse, can support the sharing of cybersecurity incidents datasets. Furthermore, anonymization techniques can be used to handle privacy concerns. Moreover, Machine Learning (ML) techniques can be used to improve IDS effectiveness. This work proposes the Sec4ML approach which supports the preparation of cybersecurity incident datasets for ML techniques using LOD practices and following FAIR principles, involving, among others, anonymization and preprocessing subprocesses, which are illustrated using public datasets.