Machine Learning-Based Cyber Threat Detection And Real-Time Network Security Enhancement

The emergence of advanced cyber-attacks targeting the latest network infrastructure has highlighted the inefficiencies of signature-based systems employed to detect such attacks, which are unable to recognize new threats due to their lack of adaptability. The current paper offers a machine learning framework for cyber threat detection and network security improvement through real-time application of streaming data analysis techniques together with ensemble learning approaches. The system is based on a hybrid framework comprising of the Random Forest and XGBoost classifiers for detection of known attacks and autoencoder-based anomaly detection method for identifying unknown threats. Feature engineering is used to extract static and behavioral features of the network traffic. In addition, the system includes an adaptive response strategy, which allows the system to dynamically modify the network security policy depending on the threat level. It was shown that the offered system is capable of providing efficient threat detection for different types of attacks, while being computationally efficient enough to be applied in a real-time scenario.