Adaptive Crypto-Steganography With Machine Learning for real-Time Network traffic

Emenike Chukwu¹; Agozie Eneh²; Nnennaya Ogbonnaya³; Precious Udechukwu⁴

¹Department of Software Engineering / Federal University of Technology, Ikot-Abasi, Akwa-Ibom State. Nigeria.
² Department of Computer Science / University of Nigeria, Nsukka. Enugu State, Nigeria
³ Department of Computer Science / University of Nigeria, Nsukka, Enugu State, Nigeria
⁴ Department of Computer Science / University of Nigeria, Nsukka, Enugu State, Nigeria

Abstract:
The increasing complexity of intrusion detection and traffic analysis methods makes it difficult to secure data transfer over open networks. Encrypted traffic patterns can nonetheless expose communication to adversary examination even though cryptography protects message content. A machine learning-enhanced adaptive crypto-steganography framework for real-time network traffic is presented in this work. Standard traffic monitoring tools like Wireshark and tcpdump are used to record live network packets.

Materials and Methods: A Python-based pipeline that uses Scapy for packet parsing, modification, and feature extraction is then used to process the packets. A supervised machine learning model built using scikit-learn is trained using key traffic variables, such as packet size distribution, inter-packet time, protocol behavior, and entropy-based measures. While lightweight encryption guarantees data secrecy before embedding, the trained model dynamically chooses the best steganographic embedding schemes and parameters during runtime. Detection probability, steganalysis resistance, robustness, and computing overhead measures are used to evaluate performance in an experimental setting with different network circumstances and traffic loads.

Results: The findings show that, in comparison to static steganographic techniques, the suggested adaptive architecture considerably lowers detectability and boosts resilience while preserving low latency appropriate for real-time deployment.

Conclusion:A machine learning-enhanced adaptive crypto-steganography framework for real-time network data was provided in this paper. Through dynamic embedding strategy selection based on real-time traffic parameters, the suggested approach increases robustness and stealth against contemporary steganalysis techniques. Deep learning-based decision models and their implementation in large-scale, high-speed network systems will be the subject of future research.

Key Word: Crypto-Steganography, Machine Learning, Network Security, Real-Time Traffic, and Covert Communication.

Adaptive-Crypto-Steganography-with-Machine-Learning-For-Real-Time-Network-Traffic-Correected-Work