Optimizing Network-Level Forensics with Optuna-Generated CNN Model

In the rapidly evolving landscape of high-speed networks (5G and beyond), accurately classifying network traffic for forensics applications poses significant challenges, particularly with respect to training loss and the computational power required for optimal hyperparameter selection. This paper explores the application of Convolutional Neural Networks (CNNs) enhanced by Optuna hyperparameter optimization for network-level forensics. The primary focus is to automate data collection and preprocessing and to reduce the false positive rate in application detection. The methodology involves leveraging CNNs with Optuna for hyperparameter tuning to achieve optimal performance. The results demonstrate improved accuracy of 2.34%, enhanced recall, a 2.31% reduction in the false positive rate, and a 17.39% decrease in the time required. This approach contributes to more sustainable practices in network security by increasing efficiency and accuracy in data processing and analysis while reducing computational resources and energy consumption.