Genetic Algorithm Based Feature Optimization for Enhanced Darknet Traffic Analysis

The detection and analysis of darknet traffic is important for enhancing network security, especially with the rise of anonymization tools like Tor and VPN that help anonymize and protect malicious user identities. Machine learning models are deployed for detection often solely focus on yielding a prediction without much importance given to preprocessing of the data. This paper presents a methodology to analyze and compare the best features of the comprehensive darknet traffic data using the CICDarknet 2020 dataset. We target the essential method of recognizing the maximum relevant features from network traffic data to yield useful models. Various machine learning models, including K-Nearest Neighbors, XGBoost, Random Forest, Decision Tree, and Logistic Regression, were evaluated for their classification performance of darknet traffic based on key features.