Optimizing Intrusion Detection Systems with Ensemble Deep Learning: A Comparative Study of RNN and LSTM Architectures
No Thumbnail Available
Date
2024-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Nowadays, due to the complexity and severity of security attacks on computer networks attackers
can launch a variety of attacks against organizational networks using a variety of methods in
order to access, modify, or delete crucial data. The rise in cyberattacks has made it necessary to
create reliable and effective intrusion detection systems (IDS) that can instantly recognize malicious
activity. IDS, which can automatically and quickly detect and categorize cyberattacks
at host and network levels, has made substantial use of machine learning techniques. Although
ML techniques like K Nearest Neighbor and Support Vector Machines have been used to building
IDSs, those systems still have a high false alarm rate and poor accuracy. Many security
researchers are integrating different machine learning approaches to protect the data and reputation
of the organizations. Deep learning algorithms have emerged as a forceful instrument
in this field and these can detect with better precision than conventional techniques. Recently,
Deep learning has become more well-known in network-based intrusion detection systems,
enhancing their efficiency in safeguarding hosts and computer networks. In the field of deep
learning, ensemble learning has appeared as a potent method that improves the performance of
single models by combining several of them. The present study employed two architectures of
recurrent neural networks (RNNs), namely simple recurrent neural networks and long shortterm
memory (LSTM), in order to investigate the possible applicability of ensemble learning
in intrusion detection systems (IDS). RNNs are suited for predicting sequential data in IDS by
identifying temporal relations in network traffic. LSTMs, which are a kind of RNN, can deal
with long-term dependencies well and help avoid vanishing gradient problem that is important
in identifying complicated intrusion model.The performance of designed model and the IDS
were evaluated using LITNET2020 publicly available dataset under performance evaluation
metrics. In multiclass classification the ensemble model fared better than LSTM, yielding accuracy
and precious 99.981% and 99.965%, respectively, whereas LSTM provided accuracy
and precious of 99.638% and 99.451 %, respectively. Additionally, the suggested ensemble
approach produced superior in multi-classification results for the various types of intrusions.
Description
Keywords
Deep Learning, Intrusion Detection System, network based Intrusion Detection System, Recurrent Neural Network