Adaptive Machine Learning for Real-Time Intrusion Detection in Networks

With the complexity of the cyber threats, the issue of developing more intelligent and flexible systems that can detect the known and unknown intrusions in real-time has become a burning matter. The paper suggests a hybrid threat detection scheme that combines the different data sampling techniques and imbalanced data treatment with the classical machine learning (ML) models in such a way that IDS performance becomes more efficient and precise to make a generic space optimised and self-learning cyber security module. Classic ML models like a Random Forest, support Vegetable Machines (SVM), and Neural Network have been extensively applied in the intrusion detection because they have the capacity to learn a pattern based on the past network traffic data. Nevertheless, they are constrained in scaling to high-dimensional data, identifying high-dimensional data or in recognizing zero-day attacks. This work implements test cybersecurity data, such as NSL-KDD etc., to gain an insight into the performance comparison of classical and quantum-enhanced models. Normalization and dimensionality reduction techniques of data preprocessing are used to provide the best possible features representation of classical and quantum classifiers. The findings reveal that classical models are performing well with the traditional forms of attacks, to identify the slightest anomalies and attack vectors complicated in nature with better accuracy and reduced false-positive rates in predetermined cases.