Secured Network Traffic (SNT) refers to communication on a network that is encoded and not subject to being intercepted or tampered with. Protecting SNT involves the implementation of multiple forms of encoding in the way of Virtual Private Networks (VPNs) for remote access and Secure Sockets Layer (SSL)/Transport Layer Security (TLS) for use over the internet. However, the encryption of traffic, rising cyber threats, and the requirement of real-time classification necessitate advanced techniques to identify more precisely and reduce false alarms. For this purpose, this research suggests the Adaptive Fitness-Dependent Optimized AutoRegressive Long Short-Term Memory (AFDO-ALSTM) model for traffic classification. Based on deep learning-based sequence modeling, this model is capable of efficiently processing encrypted traffic. The internet Security and Computer Networks Virtual Private Network (ISCX-VPN) dataset was utilized in the evaluation. First, the dataset was preprocessed through min–max normalization to normalize feature values. Principal Component Analysis (PCA) was used to eliminate irrelevant features, enhancing classification efficiency. AFDO-ALSTM was related to state-of-the-art techniques such as Residual Networks with 34 layers (ResNet34), Residual Networks with 50 layers (ResNet50), Mobile Network (MobileNet), and Self-Organizing Incremental Neural Networks (SOINN). Experimental outcomes proved that AFDO-ALSTM performed better than the existing models, as it achieved 96.94% F1-score, 97.35% recall, 98.70% accuracy, and 97.82% precision. This result highlights the effectiveness of the model in classifying SNT, increasing precision, reducing false positives, and handling encryption complexities without sacrificing real-time traffic inspection. The conclusions highlight the importance of AFDO-ALSTM in modern-day cybersecurity systems for secure and efficient network traffic classification.
Panwar et al. (Thu,) studied this question.