AI-Driven Cybersecurity Model for Real-Time Threat Detection and Prevention

As cyber threats become more dynamic and sophisticated, traditional intrusion detection systems often fall short in identifying emerging attacks in real-time. This project introduces a hybrid AI-driven cybersecurity model that combines supervised machine learning and deep reinforcement learning for adaptive, explainable, and real-time threat detection and prevention. It uses an XGBoost classifier trained on the UNSW-NB15 dataset to predict initial threats and a Deep Q-Network (DQN) agent, built with PyTorch, to make optimal security decisions based on evolving threats. The DQN interacts with real or simulated network traffic, continuously improving its policy through reward feedback.Key network features such as protocol type, port numbers, packet size, and inter-arrival time are standardized for consistent analysis. A rules-based engine supports data labeling when public datasets are insufficient. A Flask-powered dashboard provides live threat monitoring, SHAP-based model explanations, and performance insights.This modular system achieves high detection accuracy and adapts to new threats without frequent retraining. By integrating classic ML with reinforcement learning, the solution offers a future-ready cybersecurity framework that operates intelligently, efficiently, and transparently in real-time environments.