AN ADAPTIVE CYBERSECURITY THREAT PROTECTION TECHNIQUE USING MACHINE LEARNING AND BIG DATA ANALYTICS

Abstract

This research introduces a dynamic cybersecurity threat protection system that combines machine learning algorithm and big data analytics to improve detection, prediction and reaction to dynamic cyber threats. On benchmark intrusion detection data sets (CICIDS2018, UNSW-NB15), the current methodology proposes the application of data preprocessing, feature engineering, and a hybrid of Random Forest and LSTM networks. Multi-class metrics, feature importance, and detection performance are visualized on an interactive dashboard in real-time and delivered by API. Findings indicate that most of the kinds of attacks have an accuracy close to full accuracy with a few gaps still remaining in terms of detecting certain forms of threats like the SQL Injection. Adaptive learning loop of the system with continuous training and periodic retraining enables the system to remain resilient to attacks that involve zero-day threats and adversarial styles. It underscores the promise of execution driven, scalable, and explainable proposals in overcoming the shortcomings of classical signature based cybersecurity measures and at the same time underlines the importance of continuous learning and well-trained adversarial defense.