CyberNet-IDS: An AI-Driven Framework for Real-Time Cyber Attack Detection Using Deep Learning

Abstract

Rising numbers of cyber threats require real-time, intelligent IDSs that are not only based on signatures or shallow ML models. Such systems often provide poor generalisation and high false alarms, especially in cases involving more complex temporal dependencies in network traffic. Although recent developments in deep learning offer new opportunities, many of these models struggle to capture the geographical and temporal characteristics of cyber threats, making them less applicable in real-time detection environments. In response to these concerns, this study presents CyberNet-IDS, an AI-based framework for detecting cyberattacks in real time, using a simulated deep-learning hybridisation approach. We integrate a 1-dimensional CNN to abstract spatial network data and BiLSTM networks to model temporal dependencies in network traffic in our framework. Stable preprocessing methods, such as normalisation, median imputation, and an XGBoost-based feature selection process, further enhance the model. CyberNet-IDS is deployed using Apache Kafka and TensorFlow Serving, supporting the streaming of real-time packet data and classification. Through extensive experimentation on the CIC-IDS2017 dataset, we demonstrate that CyberNet-IDS outperforms traditional and existing DL-based IDSs. A well-performing detection model with a reduced false alarm rate is obtained, achieving 97.30% classification accuracy, 96.80% precision, 96.20% recall, and 96.50% F1-score. Therefore, this proposed framework addresses the fundamental issues faced by current IDS solutions and provides a scalable, deployable architecture to secure modern networks against ever-evolving cyber threats instantly.