Unified AI Models for Network Security on Edge Devices

Abstract

Rapidly identifying and mitigating security threats to reduce the impact of attacks is one of the most pressing challenges of our time. Digital threats frequently jeopardize users, often manifested as web-based, intranetwork, or spam-related attacks. The literature indicates that most studies examine each type of attack separately. This study introduces six distinct deep learning (DL) model architectures capable of detecting three attacks. The results of this study demonstrate that the Deep Neural Network (DNN) model achieved a test accuracy of 98.78% for the FWAF dataset. The CNN-LSTM model achieved 99.95% accuracy for the KDDCUP-99 dataset and 99.09% accuracy for the SMS Spam dataset. This study accurately identifies anomalous activities and anticipates potential attacks by analyzing large data sets from various data channels. Furthermore, this work improves the efficacy of existing firewalls and holds significant promise in preventing malicious factors from exploiting vulnerabilities in digital infrastructures. This study conducted experiments in a real-time environment using artificial intelligence models deployed on Jetson Nano. For malicious traffic detection, malicious traffic was created on the network with the Nmap tool, and network traffic was monitored using Wireshark, followed by classification using our model. In the Damn Vulnerable Web Application (DVWA) platform, a malicious URL classification model was used to test web attacks. For spam detection, suspicious text was retrieved from the user via a plugin running in the browser and sent to the spam classification model, and the result was returned to the user. In conclusion, DL models have the potential to detect and prevent various security threats in web applications. This study enables the analysis of normal operating behaviors in web applications and the identification of anomalous activities. This capability helps detect attempted intrusions or potential security issues indicating user behavior. Furthermore, the study facilitates the automatic detection of attacks targeting web applications and the implementation of countermeasures to mitigate such threats.