Fast learning hyper-heuristic framework for Intrusion Detection System (IDS)

Detecting cybersecurity attacks remains a challenging problem. This challenge arises from the evolving nature of attacks, a phenomenon commonly referred to as concept drift in machine learning. To address this issue, hyper-heuristic models have been identified as an effective approach. However, the various components embedded in the hyper-heuristic models have created concern about the efficiency of the model as well as its over-fitting or under-fitting of free performance. In this study, the core classifier in the hyper-heuristic model of Intrusion Detection System (IDS) is developed as parallel structure neural network (NN), which enables more controllability of reaching an optimal learning without falling into sub-optimality because of over- and under-fitting. In addition, it enables more efficiency because of reaching higher accuracy with a lower number of neurons. An evaluation of various hyper-heuristics frameworks, some of which are based on single connection NN and others based on the developed parallel connections NN provides the superiority of the latter over the former in terms of all classification metrics when lower number of neurons is used. The evaluation has been conducted on three datasets: KDD 99, NSL-KDD, and LandSat. For KDD, the reached accuracy was 97%- 99%. On the other side, we observe that the single connection has generated only an accuracy of 79% with the same number of neurons. From the computation perspective, all hyper-heuristic models have outperformed the benchmark.

Text 124130.pdf - Published Version Available under License Creative Commons Attribution. Download (1MB) Official URL or Download Paper: https://iieta.org/journals/jesa/paper/10.18280/jes...

Actions (login required)

View Item