Active-Darknet: An Iterative Learning Approach for Darknet Traffic Detection and Categorization

The darknet represents an unindexed and concealed portion of the internet often linked with illicit activities, such as trafficking drugs, weapons, and stolen data. To ensure online safety in an era of rapid technological advancement and global connectivity, the analysis and recognition of darknet traffic are crucial. This study introduces Active-Darknet, a novel machine learning-based approach employing active learning techniques to detect darknet traffic effectively. The methodology incorporates robust data preprocessing methods, including numerical encoding of categorical labels and data balancing, to enhance the representation of minority classes and improve analysis quality. In addition to leveraging active learning-based models, experiments were conducted using Deep Neural Networks (DNN), Bidirectional Long Short-Term Memory (BI-LSTM), and Flattened-DNN architectures. Among these, models based on Random Forest (RF) and Decision Tree (DT) combined with active learning achieved a high accuracy of 87%, demonstrating superior efficiency in detecting darknet traffic. The Active-Darknet approach provides robust, scalable, and effective methods for analyzing large-scale encrypted network traffic, making it valuable for cybersecurity applications. This study's results emphasize the importance of integrating machine learning and deep learning techniques with active learning frameworks to address the complexities of darknet traffic detection. Active-Darknet enhances the robustness, efficiency, and accuracy of detecting illicit activities on the darknet, contributing to the safety and security of online spaces. Keywords: Active learning, darknet traffic detection, machine learning, encrypted networks, Deep Neural Networks (DNN), Bidirectional LSTM (BI-LSTM), Random Forest (RF), Decision Tree (DT), data balancing, cybersecurity.