Image Based Approach for Classification of Network Based Security Attacks
Islam, Md Mahbub (2024)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2024092074069
https://urn.fi/URN:NBN:fi-fe2024092074069
Tiivistelmä
In the rapidly evolving field of cybersecurity, the early detection of network-based attacks has become crucial for maintaining the safety and reliability of network systems. Traditional Intrusion Detection Systems (IDS) primarily rely on signature-based detection, identifying threats only after they have occurred. This reactive approach limits the detection capabilities to known threats, making it difficult to identify new or
volving attack patterns, which can bypass traditional defenses. To address this issue, this research explores an alternative approach, employing image-based techniques for classifying network-based security attacks by transforming network traffic flows, including MQTT protocol traffic, into image representations. These representations allow the use of deep learning models, such as Convolutional Neural Networks (CNNs), to detect patterns and anomalies that may not be easily captured by conventional methods.
While image-based classification for network traffic analysis has been previously explored, this research applies the method specifically to MQTT traffic, a protocol widely used in IoT environments. It also addresses the challenge of noisy data, simulating realworld conditions where traffic may be imperfect due to transmission errors or malicious obfuscation.
The central idea of this research is to convert network traffic data into image representations, leveraging the power of deep learning models, particularly Convolutional Neural Networks (CNNs), to classify different types of network attacks. By treating network packets as visual data, this method enables the model to detect subtle patterns and anomalies that may be challenging to identify using traditional techniques. This image-based approach was tested on a custom dataset containing both normal and malicious traffic, including attacks such as brute force, Slow Denial of Service (DoS), and malformed packet attacks.
The results demonstrate that CNNs-based image classification is highly effective in classifying network traffic and detecting various forms of cyber-attacks, achieving high accuracy across different types of malicious activities. Moreover, the model’s robustness was evaluated by introducing Gaussian noise to simulate real-world conditions where network traffic is often imperfect due to transmission errors or malicious obfuscation. The findings suggest that this image-based approach not only improves the accuracy of attack detection but also increases the system’s resilience to noisy data, making it suitable for deployment in real-world network environments.
volving attack patterns, which can bypass traditional defenses. To address this issue, this research explores an alternative approach, employing image-based techniques for classifying network-based security attacks by transforming network traffic flows, including MQTT protocol traffic, into image representations. These representations allow the use of deep learning models, such as Convolutional Neural Networks (CNNs), to detect patterns and anomalies that may not be easily captured by conventional methods.
While image-based classification for network traffic analysis has been previously explored, this research applies the method specifically to MQTT traffic, a protocol widely used in IoT environments. It also addresses the challenge of noisy data, simulating realworld conditions where traffic may be imperfect due to transmission errors or malicious obfuscation.
The central idea of this research is to convert network traffic data into image representations, leveraging the power of deep learning models, particularly Convolutional Neural Networks (CNNs), to classify different types of network attacks. By treating network packets as visual data, this method enables the model to detect subtle patterns and anomalies that may be challenging to identify using traditional techniques. This image-based approach was tested on a custom dataset containing both normal and malicious traffic, including attacks such as brute force, Slow Denial of Service (DoS), and malformed packet attacks.
The results demonstrate that CNNs-based image classification is highly effective in classifying network traffic and detecting various forms of cyber-attacks, achieving high accuracy across different types of malicious activities. Moreover, the model’s robustness was evaluated by introducing Gaussian noise to simulate real-world conditions where network traffic is often imperfect due to transmission errors or malicious obfuscation. The findings suggest that this image-based approach not only improves the accuracy of attack detection but also increases the system’s resilience to noisy data, making it suitable for deployment in real-world network environments.