Balanced Efficient Lifelong Learning (B-ELLA) for Cyber Attack Detection

Rafał Kozik (TP University of Science and Technology, Poland)

Michał Choras (FernUniversitat in Hagen, Germany)

Jörg Keller (FernUniversitat in Hagen, Germany)

Abstract: This paper outlines and proposes a new approach to cyber attack detection on the basis of the practical application of the efficient lifelong learning cybersecurity system. One of the main difficulties in machine learning is to build intelligent systems that are capable of learning sequential tasks and then to transfer knowledge from a previously learnt foundation to learn new tasks. Such capability is termed as Lifelong Machine Learning (LML) or as Lifelong Learning Intelligent Systems (LLIS). This kind of solution would promptly address the current problems in the cybersecurity domain, where each new cyber attack can be considered as a new task. Our approach is an extension of the Efficient Lifelong Learning (ELLA) framework. Hereby, we propose the new B-ELLA (Balanced ELLA) framework to detect cyber attacks and to counter the problem of network data imbalance. Our proposition is evaluated on a malware benchmark dataset and we achieve promising results.

Keywords: classiffication,, cybersecurity, data imbalance, lifelong machine learning, malware detection

Categories: L.4.0