IDS-FRNN: an intrusion detection system with optimized fuzziness-based sample selection technique

Document Type

Article

Publication Date

12-1-2024

Abstract

The high intricacy rate of cyber-threats has led to the emergence of more advanced intrusion detection systems (IDSs) for identifying, detecting, and responding to potential security breaches. Over the past few years, there has been a growing trend toward utilizing machine learning (ML) models to improve the detection rate of IDSs. However, dealing with extensive or large amounts of intrusion detection (ID) data can still pose a significant challenge that may cause an IDS to experience performance degradation and the challenge of increased computational complexities (i.e., storage requirements). To alleviate this issue, this paper introduces a novel technique called IDS-FRNN which can reduce the storage space by selecting optimal and representative samples from ID dataset to enhance the effectiveness of the learning model and improve the detection rate of an IDS. IDS-FRNN initially chooses two sets of samples (i.e., representative and unrepresentative) from the data. Next, a fuzziness-based selector acquires a new group of optimized samples by using the fuzzy membership vectors outputted by a random weights neural network (RNN) from both sets of samples. The new group of samples contributes to improve the effectiveness of the IDS and achieve a trade-off between accuracy and reduction rate. The proposed technique is evaluated on three distinct attacks: Distributed Denial of Service (DDoS), DoS Hulk, and PortScan, using CIC-IDS2017 dataset. The experimental results demonstrate that IDS-FRNN significantly enhances the generalization ability of IDS, with a detection accuracy of up to 99.89% and an error rate of 0.001%. This surpasses the performance of other instance selection (IS) methods, with the ability to optimize resources and maintain high accuracy. A comprehensive comparison with existing approaches confirms IDS-FRNN’s superiority in intrusion detection and reveals valuable insights into its effectiveness.

Identifier

85205094183 (Scopus)

Publication Title

Neural Computing and Applications

External Full Text Location

https://doi.org/10.1007/s00521-024-10333-9

e-ISSN

14333058

ISSN

09410643

First Page

22789

Last Page

22803

Issue

36

Volume

36

This document is currently not available here.

Share

COinS