CADAE: Cybercrime Attack Detection using Autoencoders based on Temporal Features

Abstract

The increase in cybercrime denotes a rising trend of criminal activities conducted through digital means. This
flow is fueled by the growing dependence on technology, increased connectivity, and the explosion of online
platforms. Addressing this challenge requires enhanced cybersecurity measures to stay ahead of evolving
cyber threats. Cybercrime Attack Detection using Autoencoders (CADAE), which relies on temporal
features, poses an effective approach in which the unsupervised learning of autoencoders is advantageous
in sensitive, complex temporal patterns inherent in cybersecurity data. The CADAE approach used three
benchmark datasets: KDDCup99, CICIDS2017, and SIMARGL2022, showcasing admirable performance
in discriminating and mitigating cybersecurity threats. The KDDCup99 dataset, a pioneer in this domain,
provides a comprehensive set of labeled data, enabling the growth and assessment of intrusion detection
systems (IDSs). The CICIDS2017 dataset, designed for the assessment of network IDSs, captures a diverse
range of cyber threats, including DoS attacks and malware activity. SIMARGL2022, another significant
dataset, focuses on simulated cyber-physical systems, presenting a unique environment for assessing the
resilience of critical infrastructure against cyberattacks. The successful utilization of these benchmark
datasets underscores their effectiveness in enhancing the capabilities of intrusion detection models.