A Conditional Tabular GAN-Enhanced Intrusion Detection System for Rare Attacks in IoT Networks

Safaa Menssouri, and El Mehdi Amhoud · 2025

Read A Conditional Tabular GAN-Enhanced Intrusion Detection System for Rare Attacks in IoT Networks by Safaa Menssouri, and El Mehdi Amhoud (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-06031.txt.

Introduction

Safaa Menssouri and El Mehdi Amhoud build a guard for smart-device networks that can catch rare attacks. Such networks now run in homes, farms, and plants, yet rare attack types hide in a flood of normal traffic. A guard trained on skewed data learns the flood and misses the rare hits. Their 2025 system works in two stages to fix the skew.

First, a generative network forges fresh rows for the rare attack classes. Next, a cleanup pass lifts data quality by blending and pruning near the class edge. Tests ran on a known guard dataset with many attack kinds. The full system hit 99.9 percent overall accuracy and 80 percent on the rare classes. The team frames the gain as proof that forged data, used with care, can teach guards to see what they once missed.

Catalog record

CREATORSafaa Menssouri, and El Mehdi Amhoud
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE22 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-06031

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Conditional Tabular GAN-Enhanced Intrusion Detection System for Rare Attacks in IoT Networks Safaa Menssouri, and El Mehdi Amhoud Abstract Internet of things (IoT) networks, boosted by 6G technology, are transforming various industries. However, their widespread adoption introduces significant security risks, particularly in detecting rare but potentially damaging cyber-attacks. This makes the development of robust intrusion detection systems (IDS) crucial for monitoring network traffic and ensuring their safety. Traditional IDS often struggle with detecting rare attacks due to severe class imbalances in IoT data. In this paper, we propose a novel two-stage system called conditional tabular generative synthetic minority data generation with deep neural network (CTGSM-DNN). In the first stage, a conditional tabular generative adversarial network (CTGAN) is employed to generate synthetic data for rare attack classes. In the second stage, the SMOTEENN method is applied to improve dataset quality. The full study was conducted using the CSE-CIC-IDS2018 dataset, and we assessed the performance of the proposed IDS using different evaluation metrics. The experimental results demonstrated the effectiveness of the proposed multiclass classifier, achieving an overall accuracy of 99.90% and 80% accuracy in detecting rare attacks. {safaa.menssouri, elmehdi.amhoud}@um6p.ma I Introduction The proliferation of internet of things (IoT) networks, coupled with the emergence of 6G technology [ 1 ] , are transforming numerous aspects of modern life, from smart homes and cities to industrial automation and healthcare systems. IoT networks comprise a vast array of interconnected devices that collect, exchange, and analyze data to facilitate intelligent decision-making and automation. This interconnected ecosystem enhances efficiency, productivity, and convenience, driving the rapid adoption of IoT technologies across various domains. However, the widespread deployment of IoT devices within the high-speed, low-latency framework of 6G introduces substantial security challenges. The heterogeneous nature of these networks, combined with their large-scale and distributed architecture, makes them vulnerable to a variety of cyber threats [ 2 ] . Traditional security measures are often inadequate for addressing the unique requirements and vulnerabilities inherent in IoT environments. Thus, robust and adaptive security solutions, such as intrusion detection systems (IDS) [ 3 ] , are essential for monitoring network traffic, detecting anomalies, and safeguarding against potential security breaches. Although there are different types of IDS, in our research work we are interested in anomaly-based IDS (AIDS) [ 4 ] . These systems aim to detect both known and unknown attacks, providing a more comprehensive detection capability than other IDS. One of the significant challenges in developing effective anomaly-based IDS is the issue of imbalanced data. In many real-world scenarios, the number of normal instances vastly outweighs the number of anomalous ones, creating a class imbalance problem. This imbalance is particularly problematic when dealing with rare classes of attacks. Rare attacks are a type of security breach that occur infrequently and have a low number of instances, making them less familiar and more challenging to detect. They often account for only a small fraction of the normal data, and when using a single machine learning (ML) model for classification, the classifier tends to favor the majority class of data and misclassify the rare-class attack data. Despite their rarity, these attacks can indicate more sophisticated and potentially damaging threats. As a result, developing effective techniques to detect rare intrusion events is crucial for maintaining the security and integrity of IoT networks. In this work, we aim at enhancing the detection capabilities for rare attacks by exploring novel approaches and leveraging advanced ML techniques. While several previous studies have focused on using generative adversarial networks (GANs) for data augmentation to address the challenge of imbalanced datasets, our approach is distinguished by the use of conditional tabular GAN (CTGAN). Unlike traditional GANs, which can struggle with discrete data characteristics, CTGAN excels at capturing the nuances of rare attack classes. The main contributions of this paper are summarized as follows: • We develop a novel two-stage methodology that first employs CTGAN to generate synthetic data for rare attack classes, followed by the SMOTEENN method to enhance the detection of attacks in IoT networks. • We validate our proposed system on one of the largest public datasets (CSE-CIC-IDS2018) for IDS, and we conduct comprehensive comparisons with existing methods. • The experiments demonstrate that our proposed solution significantly improves the detection accuracy of rare attack instances, achieving up to 80% accuracy for rare attacks while maintaining 99.90% overall classification accuracy. The remainder of the paper is organized as follows: In section II, we introduce the related work. In section III, we detail the architecture of our proposed model along with a description of the dataset. In section IV, we present our simulation results, and we discuss the observations and findings. Finally, in section V, we conclude and outline our perspectives. II Related work Intrusion detection systems play a crucial role in safeguarding IoT networks. Numerous studies have focused on developing effective IDS for IoT networks, employing a variety of techniques ranging from traditional signature-based methods to more advanced ML approaches. Notably, supervised learning algorithms, such as k-nearest neighbors (KNN), naïve bayes, and support vector machines (SVM) [ 5 ] , have been extensively studied for their effectiveness in detecting anomalies. In [ 6 ] , the authors conducted several experiments to evaluate the efficiency and performance of various ML classifiers, such as random forest, random tree, decision table, naive Bayes, and Bayes network. All the tests were conducted using the KDD dataset. The experiments demonstrated that there is no single ML model that can handle efficiently all the types of attacks. Furthermore, deep learning (DL) techniques, such as convolutional neural networks (CNN) or long short-term memory (LSTM) [ 7 ] are widely used in IDS . The use of a single ML model has inherent limitations [ 8 ] . Thus, in recent years, various learning algorithms have been combined to enhance performance of IDS [ 9 ] . For instance, in [ 10 ] the authors proposed an IDS that combines the powerful learning ability of LSTM in time series data with CNN ’s ability to extract spatial features. The model was trained using KDD CUP99, NSL-KDD, and UNSW-NB15 classic datasets, the results show good convergence and performance. Despite these advancements, a significant challenge remains in dealing with imbalanced datasets, particularly regarding rare attacks detection. Traditional and even some advanced ML models often struggle to accurately detect rare attacks due to their scarcity in the training data. Xu et al. [ 11 ] presented a recurrent neural network based IDS . Their approach, tested on the NSL-KDD and KDD Cup’99 datasets, demonstrated promising results compared to other methodologies. However, it has a limitation in detecting minority attack classes, such as U2R and R2L, resulting in lower detection rates for these specific classes. Other works, such as [ 12 ] , have utilized ensemble learning with feature selection technique to enhance IDS performance on the CSE-CIC-IDS2018 dataset. While the overall detection accuracy has been improved, rare attack classes like SQL Injection and Brute Force were not well classified. Additionally, the infiltration attack class showed lower performance, revealing challenges in detecting these types of attacks. Moreover, in [ 13 ] the authors proposed a Bagging-DNN-based IDS that addresses class imbalance by incorporating class weights and leveraging deep neural networks

Read the complete text