Authors:
Amine Hattak
1
;
2
;
Fabio Martinelli
1
;
Francesco Mercaldo
3
;
1
and
Antonella Santone
1
Affiliations:
1
Institute for Informatics and Telematics, National Research Council of Italy (CNR), Pisa, Italy
;
2
La Sapienza, University of Rome, Rome, Italy
;
3
University of Molise, Campobasso, Italy
Keyword(s):
Internet of Things, Network Traffic Classification, Deep Learning, Network Intrusion Detection, Explainable AI.
Abstract:
In an era marked by escalating cyber threats, ensuring the security of interconnected devices and networks within the Internet of Things (IoT) landscape is imperative. This paper addresses this pressing concern by delving into network security, focusing on the classification of network traffic through the lens of deep learning techniques. Our study presents a deep learning-based approach customized for network traffic classification in the IoT domain, based on image analysis. Crucially, to enhance the interpretability and the transparency in our model’s decisions, we integrate GradCAM (Gradient-weighted Class Activation Mapping), a technique that illuminates the salient regions of input images contributing to the model’s predictions. By leveraging GradCAM, we provide deeper insights into the decision-making process, enabling better understanding and trust in our approach. We evaluate the effectiveness of our methodology using the TON IoT dataset, consisting of 10 network traces categ
orized into various vulnerability scenarios and trusted applications. Our findings reveal a remarkable accuracy of 99.1%, demonstrating the potential of our approach in fortifying network security within IoT environments. Moreover, the utilization of GradCAM empowers stakeholders with valuable insights into the inner workings of the model, further enhancing its applicability and trustworthiness.
(More)