MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641075343 A) filed by Hindusthan Institute Of Technology on June 18, 2026, for Deep Learning Based Real Time Network Intrusion And Zero Day Attack Detection System.
Inventors include Dr. C. Natarajan; K. R. Kannan; Dr. B. Paulchamy; Dr. S. Kavitha; P. Revathi; S. Manikandan; Y. Vijay; and K. Yeshwanth.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: The rapid growth of interconnected networks and digital services has significantly increased the risk of cyber attacks, making network security a critical concern for organizations and individuals. Traditional intrusion detection systems often rely on predefined signatures and struggle to identify unknown or zero-day attacks. To address this challenge a deep learning-based approach for zero- day attack detection in network traffic using an unsupervised Auto encoder model trained on the NSL- KDD dataset, a widely accepted benchmark for network intrusion detection. The proposed system leverages there. construction capability of the Auto encoder to learn normal network behavior during training. The model is trained exclusively on normal traffic instances, allowing it to reconstruct legitimate patterns with minimal error. Any significant deviation from the learned normal behavior measured through reconstruction error (Mean Squared Error) is flagged as an anomaly, potentially indicating a zero day attack. The reconstruction error threshold is determined statistically using mean+ 3 xstandard deviation from the training set's normal traffic reconstruction errors, ensuring high sensitivity to unseen malicious patterns. The system is implemented using· Tensor Flow/Keras for model development, scikit- learn for pre processing (Min Max scaling and categorical encoding) and Stream lit for the deployment of an interactive web- based dashboard. The model demonstrates effective anomaly detection by achieving high reconstruction error on attack instances while maintaining low error on normal traffic. ~ The system provides an intuitive, cyber- themed user interface with matrix rain f background,. glitch effects and real-time alerts, making it suitable for both educational ~ and practical cyber security applications. By focusing on unsupervised learning, the ~ proposed solution addresses the critical limitation of supervised methods in zero-day E scenarios, offering a proactive defense mechanism against unknown network threats.
Disclaimer: Curated by HT Syndication.
