Professional Documents
Culture Documents
Multi Level Deep Learning Model For Network Anomal
Multi Level Deep Learning Model For Network Anomal
8–19
Article history: The increasing use of internet-based solutions and services in private and corporate sectors
Received: 9 /11/2023 has resulted in a significant increase in personal internet involvement. This shift, concurrently
Rrevised form: 3 /12/2023 exposed a heightened susceptibility to potential vulnerabilities, given the ability of malicious
actors to exploit external networks, network services, or corporate infrastructures utilized for
Accepted : 15 /12/2023
personal purposes. In recent times, there has been considerable interest in harnessing deep
Available online: 30 /12/2023 learning methodologies for enhancing cybersecurity, owing to their utilization of sophisticated
learning algorithms for addressing pertinent online security challenges. Machine Learning
Keywords: (ML) and Deep Learning (DL) paradigms have been extensively applied across diverse
dimensions of cybersecurity, encompassing tasks such as vulnerability assessment, malware
Anomaly detection classification, spam detection, and spoofing identifica tion.
In this paper, for hierarchical intrusion detection is proposed a novel multi-stage approach, The
Machine Learning proposed system comprises two distinct classification modalities: multi-class classification and
binary classification, contingent upon the nature of the attack within the dataset. the KDD99
Deep Learning dataset was leveraged to assess the classification performance of the proposed model. Both
classification approaches involve main preprocessing steps, such as feature selection, feature
cybersecurity normalization, building a Convolutional Neural Network (CNN) classifier apply on KDD99
dataset, deploying the CNN classifier for anomaly detection.
https:// 10.29304/jqcsm.2023.15.41346
1. Introduction
The issue of network security has become increasingly significant due to the exponential rise of computer network
usage and the proliferation of billions of applications operating within this framework. Irrespective of individual or
organizational users, contemporary hackers have identified more avenues to exploit vital information [1]. Hence, the
reliable detection of assaults and anomalies in networks becomes increasingly crucial for intrusion detection systems
(IDSs). It is crucial to priorities developing resilient detection techniques on Intrusion Detection Systems (IDSs) to
effectively safeguard against the continuously expanding range of harmful behaviors [1, 2]. Network anomalies can
arise from various factors, including network overload, device malfunctions, misconfigurations, malicious actions, or
network attacks that intercept and analyze standard network services [3, 4].
In the contemporary age, safeguarding digital systems from cyber threats is a paramount concern for IT
infrastructures and the broader technological landscape. The persistent endeavors to infiltrate national and
organizational network structures underscore the insufficiency of traditional firewall and antivirus protections
against advanced digital attacks. This insufficiency leads to potential vulnerabilities in systems. Recent times have
∗Maythem S. Derweesh
borne witness to numerous instances of organizations falling victim to cyber assaults and data breaches, culminating
in substantial financial losses reaching millions of dollars and the exposure of customer information. Notable
examples of these incidents include Twitter, SolarWinds, and Finastra. Projections indicate a staggering surge from
$3 trillion to $10.5 trillion in costs attributed to cybercrime, positioning it as an unprecedented "transfer of economic
wealth" [5].
The rapidly expanding network has brought efficiency, convenience, and a growing demand for high-quality service.
Conventional machine learning techniques, including Bayesian [6], Support Vector Machines [7], Decision Trees [8]
and Logistic Regression [9], have been extensively employed in network intrusion detection systems, and other
similar methods. These strategies have yielded favorable outcomes. Nevertheless, these techniques are inadequate
for handling extensive and complex data sets, and are incapable of addressing the issue of diminished classification
accuracy caused by their susceptibility to outliers and noise. Simultaneously, traditional machine learning methods
have struggled to match user demands due to the ongoing advancement of digital technology and the growing variety
of cyberattacks.
There exist two distinct categories of Intrusion Detection Systems (IDSs). There are two types of attack detection
methods: misuse detection, which identifies assaults by comparing them to known patterns, and anomaly detection,
which identifies aberrant attacks by comparing them to regular usage patterns. While the detection of bad using
makes the recognition of not-familiar attacks hard, the detection of abnormal states has the ability to detect the new
attacks; however, the detection of abnormal attacks can generate a lot of false alarms because of the deficiencies in
accurately detecting the normal patterns. Deep learning (DL) is a method to solve this limitation by developing unique
attributes with a sophisticated neural network. Utilizing Deep Learning in Intrusion Detection Systems can mitigate
the limitations of IDS. Machine Learning (ML) and Deep Learning (DL) autonomously acquire knowledge of an
intrusion set and identify regular usage patterns, hence minimizing the occurrence of false alerts.
Anomaly based system for intrusion detection is a solution for cyber security that aims to identify cyber attacks by
monitoring and classifying the system and network activities as normal or abnormal procedures. The goal of these
systems is to reach a balance between reducing false alarms and increasing the detection rate [10]. Anomaly detection,
including conntinous monitoring and analysis of typical network patterns, allows for the detection of potential threats
and the identification of deficiencies from the known standards. The anomaly events are considered significant
because they refer to atypical events and may be critical, such as abnormal flows of traffic indicating of ongoing
assaults or illicit transmission of data activities. Anomaly could be categorized as three distinct kinds: point, collective,
and contextual. Each type of attack corresponds to specific security breaches, including Denial of Service (DoS), Probe,
User to Root (U2R), and Remote to Local (R2L) attacks [11].
Efficiently spotting and categorizing these anomalies necessitates the adaptation of network intrusion detection
systems (NIDS) to dynamic network settings, encompassing novel protocols and behaviors. Various approaches, such
as statistical, knowledge-based, and machine-learning techniques, have been harnessed in anomaly-based NIDS.
However, research challenges need addressing to enhance Performance and align these techniques with the ever-
evolving characteristics of network data [12].
The frequency of internet attacks is on the rise, and in response to these threats, two fundamental methods are
employed to safeguard information security with the principles of CIA (confidentiality, integrity, availability) [13] in
mind: Identification based on signatures and detection based on anomalies. Signature-based methods rely on specific
databases specifically created to identify and categorize potential risks. Research has demonstrated the effectiveness
of this approach, but it necessitates continuous updates to the databases and the processing of new attack data.
Furthermore, even with up-to-date databases, they remain susceptible to zero-day attacks, which are previously
unseen threats that databases cannot defend against [14].
This paper proposes a hierarchical intrusion detection system consisting of multiple stages to determine whether
data is normal or attack. Subsequently, data identified as an attack is classified into specific attack types. The proposed
system is evaluated using the KDD99 dataset.
In our proposed model, binary classification is implemented as one stage, while multiple classification serves as a
separate stage. This approach reduces the volume of data processed in the second stage (multi-class), resulting in
more efficient processing and significantly reduced classification time.
in Section 2. related works are presented, in Sections 3 describe Network Anomaly Detection, in section 4 and 5,
respectively. We develop our CNN based ADS model and test our model and in Section 5 the conclusion.
Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19 9
2. Related Works
Numerous research endeavors and studies have been dedicated to enhancing the precision and detection rates of
identifying anomalous network traffic through the application of diverse technologies.
In 2019, Xiao et al. utilized a Convolutional Neural Network (CNN) within the framework of Intrusion Detection
Systems (IDS). Their objective was to extract critical features from data that had undergone dimensionality reduction.
The CNN model was employed to process and interpret the dimensionality-reduced data, ultimately enabling the
detection of network intrusions [15]. In 2019, Yang et al, used (CNN) network that upgraded to detect wireless
network attacks, the changes in (CNN) framework aim to increase the efficiency of detecting the wireless attacks [16].
In 2019, Lin et al, proposed combining LSTM and AM to improve pattern recognition. LSTM, known for its memory,
helps remember network traffic patterns. The hierarchical neural network-based LSTM method combines current
data with historical knowledge to improve classification outcomes. Deep learning uses LSTM to store and keep
information, like the human brain. The Attention Mechanism (AM) replicates the brain's ability to focus on crucial
details, helping the model detect significant data patterns. LSTM and AM increase network performance by identifying
and interpreting complex data patterns [17]. In 2020, Karatas et al. used six machine learning methods to examine
the CICIDS2018 dataset in 2020. The machine learning methods mentioned were adaptive boosting (AdaBoost),
decision tree (DT), random forest (RF), k-nearest neighbours (KNN), gradient boosting (GB), and linear discriminant
analysis. The not equal of assault types in the dataset was addressed via synthetic minority oversampling (SMOTE).
Synthetic samples for underrepresented attack categories were generated using SMOTE, improving minority class
case detection [18]. In 2020, Mohamed et al, study the CICIDS2018 and Bot-IoT datasets, they tests sevent methods
to deep learning, used DNN, CNN, RNN, DBN, DBM, RBM, and deep autoencoder, this study test the efficiency of
modern deep learning techniques inintrusion detection [19]. In 2020, Zhiquan et al, improved Intrusion Detection
Systems using CNN and Adaptive Synthetic Sampling ADASYN. The IDS uses a uniform method to enhance the
precision and dependability of its abilities to improve its network attack detection accuracy [20]. In 2020, Kaiyuan et
al, created IDS by combines hybrid sampling with CNN and Bidirectional Long Short-Term Memory. The hybrid
technique uses CNN and BiLSTM to monitor and analyze network operations to improve network intrusion detection
using both (CNN) and (BiLSTM) techniques [21]. In 2020, Feng et al, used (LSTM-RNN) to develop full system to
extract the complex attacks. their methodology included upgrade full structure that smoothly combine among data
preparing, features extraction, training and detection. the system designed and engineered carefully to effectively deal
with the various information channels, resulting to create extensive wide work to detect attacks. [22]. In 2020,
Jahanzaib et al, this paper presents a security protection framework designed specifically to secure the control layer
of a software defined network. For adept and prompt intrusion detection system (IDS), this platform employs the
capabilities of (LSTM) and (CNN) methodologies [23]. In 2020, Jiyeon et al, the researchers made a study using KDD99
and CICIDS2018 datasets, that contained a wide range of attacks types. however, they mainly focus on defeat the
Denial of Service (DoS) attacks. the choice of using these datasets inspired from the high international frequency of
Distributed Denial of Service (DoS) attack incidents, which offered of a lot of chances for training and evaluate the
proposed Convolutional Neural Network (CNN) architecture. The CNN architecture chosen because of it is efficiency
in extracting the high value information from the extensive datasets via it is convolutional layers [24]. In 2021,
Anzhelika et al, performed a study using the two datasets KDD99 and CICIDS2018. they used the structure of U-Net
with temporal convolutional network (TCN) to identify the network threats. to enhance the accuracy, they combined
the TCN and LSTM models, that resulted to enhanced the performance. TCN played a critical part in effective
processing time series data by using the wide convolutions, that resulted to increase the convolution range and lead
to higher outputs [25]. In 2021, Yakubu et al, proposed approach named (BiDLSTM) to intrusion detection system
(IDS) to identify the attacks Remote-to-Local (R2L) and User-to-Root (U2R) accurately. during execute their model,
they achieved high level of accuracy compared to using (LSTM) traditional approach [26]. In 2021, FatimaEzzahra et
al, used Principal Component Analysis (PCA) on the dataset employing a mutual information technique. The goal was
to reduce the number of dimensions and show important characteristics. The dataset performed Principal Component
Analysis (PCA) using a mutual information technique to reduce dimensionality and find important characteristics.
Upon finishing the feature extraction procedure, they proceeded to create a detection methodology for attacks in the
dataset using (LSTM) [27]. In 2021, Satish Kumar et al, study the changing environment of network attacks and the
importance (IDS) in ensuring network security.
The article reviews research trends in Network-based Intrusion Detection Systems (NIDS), covering approaches and
commonly used datasets. It analyzes the popularity of various NIDS techniques based on citations and publications
10 Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19
over a 15-year period [28]. In 2022, Theyazn et al, Employing Convolutional Neural Networks (CNN) and a
combination of CNN and Long Short-Term Memory (CNN-LSTM), the focus was on fortifying the security of
autonomous vehicles from potential intrusions. undertook the task of training and assessing these techniques using
an authentic dataset sourced from an autonomous vehicle network. dataset encompassed an array of attack types,
including spoofing, flooding, and replay attacks, alongside legitimate packets [29]. In 2023, Tianhao Hou and
colleagues proposed a approach for Network Intrusion Detection (NID) by integrating the log-cosh conditional
variational autoencoder (LCVAE) with a convolutional bi-directional long short-term memory neural network
(LCVAE-CBiLSTM) utilizing deep learning (DL) methodologies. This method enables virtual samples to acquire
distinct threat data, hence augmenting the possible characteristics related to unbalanced assault types. In order to
consider both the temporal and spatial elements of assaults, a hybrid feature extracting model is proposed. This model
mixes (CNN) with (BiLSTM) networks [30]. In 2023, Jiaming Song et al, introduce a network intrusion detection
model, denoted as CSK-CNN, which is built upon a two-layer convolutional neural network (CNN) and employs the
Cluster-SMOTE + K-means algorithm to handle imbalanced datasets. The CSK-CNN model combines Cluster-SMOTE,
a synthetic minority oversampling technique based on clustering, with a K-means-based under sampling approach.
This two-layer network not only enables the identification of abnormal network traffic but also facilitates the
classification of such traffic into specific attack categories [31].
We want intrusion detection systems to possess the capability to not only detect the presence of an attack but also
accurately classify the specific type of attack. Consequently, certain researchers have utilized Convolutional Neural
Networks (CNN) to build a multi-class intrusion detection system. Nevertheless, their investigations revealed that,
akin to the real-world distribution of attacks, the dataset exhibits significant variation in the amount of data across
different classes. Hence, the rate of recognition is unsatisfactory for attack categories that have a limited amount of
training data. The strategy presented in this research aims to address the limitations identified in the previous
investigations. The previous works are summarized in Table 1.
3. Network attacks
Network security's role is to protect digital information, preserving data confidentiality, integrity, and resource
availability [30]. In simple terms, a threat or attack encompasses any entity with harmful attributes aimed at
undermining a network. Weak network design, user negligence, and misconfiguration of software or hardware can
render a network susceptible to such attacks [32].
Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19 11
4. Proposed Methodology
Figure 2 illustrates the overall architecture of the system's framework. During the dataset preprocessing phase, we
executed a range of procedures and operations on the data. During the second stage, the characteristics of each data
sample were standardized prior to being fed into the CNN model. During the third step, the most optimal
characteristics are chosen. The CNN model underwent training and testing in the fourth step, utilizing the KDD99
dataset. Ultimately, we employed the Convolutional Neural Network (CNN) model for the purpose of anomaly
detection.
12 Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19
The model trained by Lenovo laptop with hardware capabilities of 24 Gigabyte of RAM, Core i7 2.30GHz CPU, and 64-
bit windo ws 1 p rofe ssiona l syste m. Al l e xpe rime nta l re sul ts conducte d python 3 e nvi ronme nt.
The main steps of the proposed system are summarized as follows:
Preprocessing
In order to get better results, the data goes through some necessary preprocessing procedures such as feature
normalization and feature selection .Consequently, protocol types are converted into integer numerical values. Label
encoding is particularly useful for unordered and case-sensitive categorical data, as it assigns each category an integer
value in an alphabetical manner.
(1)
- Feature Selection
Feature engineering is a crucial preprocessing step that offers several benefits, such as reducing computing costs,
saving time, and improving model performance. Additionally, it involves determining the most influential feature
selection strategy from various types of feature selection strategies.
CNN model has feature selection inside his architecture, and we can add additional feature selection step before input
data in the CNN model and we use (Information Gain) to decrease the number of features and reduce the process time.
Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19 13
- Dataset balancing
We employed the SMOTE technique to achieve dataset balancing. SMOTE is an oversampling method that creates
synthetic samples specifically for the minority class. This technique mitigates the problem of overfitting caused by
random oversampling. This approach focuses on the feature space. in order to produce novel instances by means of
interpolation between closely located positive instances.
The heart of our proposed methodology lies in the anomaly detection and classification stage, where Convolutional
Neural Networks (CNNs) are employed to identify and categorize network intrusions.
The current stage is segmented into the subsequent phases:
- CNN Architecture
To commence, the initial step involves the development of a Convolutional Neural Network (CNN) structure
specifically designed for the purpose of network intrusion detection. Figure 3 illustrates that the architecture consists
of several convolutional layers, followed by pooling layers. These layers allow the model to autonomously acquire
structured features from the preprocessed network traffic data.
The Conv1 and Conv2 Layers are 1D convolutional layers that are designed to learn local patterns or features in the
input data, the choice of the number (64) of filters are based on empirical experimentation and to reach the best
classification results.
- Model training
The dataset split into training data (80%) and testing data (20%). The CNN model is trained on a labeled dataset that
includes both normal network traffic and various types of intrusions. During training, the model learns to differentiate
between normal and anomalous patterns by adjusting its internal parameters through backpropagation
5. Results
Proposed method achieved good results, the accuracy for Binary classifying (0.9933) and for multi-classifying
(0.9983) as presented in Table 3.
Table 3. illustrate the Performance of the classifiers on Binary and multi-class classification.
Stage 1
Binary
Stage 2
Multi-class
In order to further verify the effectiveness of the anomaly detection model proposed in this paper, a performance
comparison experiment is carried out in this part. Under identical experimental conditions, we evaluated the
performance of well-known machine learning methods, including RNN, LSTM, and other state-of-the-art intrusion
detection models, on the given dataset. The performance comparison is displayed in Table 4. The bold words signify
the assessment criteria of the suggested model.
6. Conclusion
We have created a convolutional neural network (CNN) model to detect anomalies in networks, specifically utilizing
the KDD99 dataset. The KDD99 dataset consists of four attack categories, namely DoS (Denial of Service), U2R (User
to Root), R2L (Remote to Local), and Probing. The majority of KDD studies that utilize deep learning techniques have
focused on performing binary classifications to differentiate between normal and attack instances within the full
dataset. These studies have also conducted multiclass classifications to differentiate the four groups in KDD. Our CNN
model has been created taking into account the number of convolutional layers and the dimensions of the kernel. To
evaluate our model, we created multiple scenarios, considering hyperparameters such as the number of filters, the
amount of convolutional layers, and the kernel size described before. We conducted binary and multiclass
classifications for each situation and subsequently recommended the ideal scenarios that exhibit superior
performance. The empirical findings indicate that our suggested approach outperforms previous studies in terms of
accuracy, both in binary and multiclass classifications. Nevertheless, the model presented in this article exhibits
certain limitations, requiring substantial resources and featuring a rather large number of parameters. significantly
study should explore methods for minimizing latency in deep learning-based intrusion detection systems in order to
significantly decrease the computational time required.
Acknowledgements
The authors are thankful to the Department of Computer Science, College of Science, Mustansiriyah University
(https://uomustansiriyah.edu.iq/e-newsite.php), for supporting this work.
References
[1] Chen, L., et al. Zyell-nctu nettraffic-1.0: A large-scale dataset for real-world network anomaly detection. in 2021 IEEE International Conference
on Consumer Electronics-Taiwan (ICCE-TW). 2021. IEEE.
[2] Fernandes, G., et al., A comprehensive survey on network anomaly detection. Telecommunication Systems, 2019. 70: p. 447-489.
[3] Bhattacharyya, D.K. and J.K. Kalita, Network anomaly detection: A machine learning perspective. 2013: Crc Press.
[4] hashim, h.b., Challenges and Security Vulnerabilities to Impact on Database Systems. Al-Mustansiriyah Journal of Science, 2018. 29(2): p. 117-
125.
[5] Almulla, K., Cyber-attack detection in network traffic using machine learning. 2022.
[6] Ren, X., W. Jiao, and D. Zhou, Intrusion detection model of weighted navie bayes based on particle swarm optimization algorithm. Computer
Engineering and Applications, 2016. 52(7): p. 122-126.
[7] Sahu, S.K., et al., An SVM-based ensemble approach for intrusion detection. International Journal of Information Technology and Web
Engineering (IJITWE), 2019. 14(1): p. 66-84.
[8] Ahmim, A., et al. A novel hierarchical intrusion detection system based on decision tree and rules-based models. in 2019 15th International
Conference on Distributed Computing in Sensor Systems (DCOSS). 2019. IEEE.
[9] Ioannou, C. and V. Vassiliou. An intrusion detection system for constrained WSN and IoT nodes based on binary logistic regression. in
Proceedings of the 21st ACM International Conference on Modeling, Analysis and Simulation of Wireless and Mobile Systems. 2018.
[10] Mishra, S., et al., Swarm intelligence in anomaly detection systems: an overview. International Journal of Computers and Applications, 2021.
43(2): p. 109-118.
[11] Rana, S., Anomaly Detection in Network Traffic using Machine Learning and Deep Learning Techniques. Turkish Journal of Computer and
Mathematics Education (TURCOMAT), 2019. 10(2): p. 1063-1067.
[12] Khan, W. and M. Haroon, An unsupervised deep learning ensemble model for anomaly detection in static attributed social networks. International
Journal of Cognitive Computing in Engineering, 2022. 3: p. 153-160.
[13] Haqi Al-Tai, M., B.M. Nema, and A. Al-Sherbaz, Deep Learning for Fake News Detection: Literature Review. Al-Mustansiriyah Journal of
Science, 2023. 34(2): p. 70-81.
[14] Khan, A.R., et al., Deep learning for intrusion detection and security of Internet of things (IoT): current analysis, challenges, and possible
solutions. Security and Communication Networks, 2022. 2022.
[15] Xiao, Y., et al., An intrusion detection model based on feature reduction and convolutional neural networks. IEEE Access, 2019. 7: p. 42210-
42219.
[16] Yang, H. and F. Wang, Wireless network intrusion detection based on improved convolutional neural network. Ieee Access, 2019. 7: p. 64366-
64374.
[17] Lin, P., K. Ye, and C.-Z. Xu. Dynamic network anomaly detection system by using deep learning techniques. in Cloud Computing–CLOUD
2019: 12th International Conference, Held as Part of the Services Conference Federation, SCF 2019, San Diego, CA, USA, June 25–30, 2019,
Proceedings 12. 2019. Springer.
[18] Karatas, G., O. Demir, and O.K. Sahingoz, Increasing the performance of machine learning-based IDSs on an imbalanced and up-to-date dataset.
IEEE access, 2020. 8: p. 32150-32162.
[19] Ferrag, M.A., et al., Deep learning for cyber security intrusion detection: Approaches, datasets, and comparative study. Journal of Information
Security and Applications, 2020. 50: p. 102419.
18 Maythem S. Derweesh , Journal of Al-Qadisiyah for Computer Science and Mathematics VOL. 15(4) 2023 , PP COMP. 8–19
[20] Hu, Z., et al., A novel wireless network intrusion detection method based on adaptive synthetic sampling and an improved convolutional neural
network. IEEE Access, 2020. 8: p. 195741-195751.
[21] Jiang, K., et al., Network intrusion detection combined hybrid sampling with deep hierarchical network. IEEE access, 2020. 8: p. 32464-32476.
[22] Jiang, F., et al., Deep learning based multi-channel intelligent attack detection for data security. IEEE transactions on Sustainable Computing,
2018. 5(2): p. 204-212.
[23] Malik, J., et al., Hybrid deep learning: An efficient reconnaissance and surveillance detection mechanism in SDN. IEEE Access, 2020. 8: p.
134695-134706.
[24] Kim, J., et al., CNN-based network intrusion detection against denial-of-service attacks. Electronics, 2020. 9(6): p. 916.
[25] Mezina, A., R. Burget, and C.M. Travieso-González, Network anomaly detection with temporal convolutional network and U-Net model. IEEE
Access, 2021. 9: p. 143608-143622.
[26] Imrana, Y., et al., A bidirectional LSTM deep learning approach for intrusion detection. Expert Systems with Applications, 2021. 185: p. 115524.
[27] Laghrissi, F., et al., Intrusion detection systems using long short-term memory (LSTM). Journal of Big Data, 2021. 8(1): p. 65.
[28] Kumar, S., S. Gupta, and S. Arora, Research trends in network-based intrusion detection systems: A review. IEEE Access, 2021. 9: p. 157761-
157779.
[29] Aldhyani, T.H. and H. Alkahtani, Attacks to automatous vehicles: A deep learning algorithm for cybersecurity. Sensors, 2022. 22(1): p. 360.
[30] Hou, T., et al., A Marine Hydrographic Station Networks Intrusion Detection Method Based on LCVAE and CNN-BiLSTM. Journal of Marine
Science and Engineering, 2023. 11(1): p. 221.
[31] Song, J., et al., CSK-CNN: Network Intrusion Detection Model Based on Two-Layer Convolution Neural Network for Handling Imbalanced
Dataset. Information, 2023. 14(2): p. 130.
[32] Van, N.T. and T.N. Thinh. An anomaly-based network intrusion detection system using deep learning. in 2017 international conference on
system science and engineering (ICSSE). 2017. IEEE.
[33] Ferrag, M.A., et al., Rdtids: Rules and decision tree-based intrusion detection system for internet-of-things networks. Future internet, 2020. 12(3):
p. 44.
[34] Mahmood, H.A., Network Intrusion Detection System (NIDS) in Cloud Environment based on Hidden Naïve Bayes Multiclass Classifier. Al-
Mustansiriyah Journal of Science, 2018. 28(2): p. 134 - 142.
[35] Damasevicius, R., et al., LITNET-2020: An annotated real-world network flow dataset for network intrusion detection. Electronics, 2020. 9(5):
p. 800.
[36] Labonne, M., Anomaly-based network intrusion detection using machine learning. 2020, Institut polytechnique de Paris.
[37] Li, R., et al., GTF: An Adaptive Network Anomaly Detection Method at the Network Edge. Security and Communication Networks, 2021. 2021:
p. 1-12.