Deep Q-Network Achieves 99.72% Accuracy in Cloud Intrusion Detection
So, there’s this new study on arXiv (2608.12190) that proposes a cool cyber defense system for cloud setups. It uses reinforcement learning with something called a Deep Q-Network (DQN) to create strategies against evolving cyber threats. They built it using the CICIDS2017 dataset and tested it on UNSW-NB15, incorporating steps like data prep, feature engineering, and adaptive learning. When they compared the DQN to other models like decision trees and random forests, it really excelled, hitting an accuracy of 99.72% and a precision of 99.68%. The false positive and negative rates were super low at 0.31% and 0.35%, respectively, and it has a quick detection time of 15 ms. This framework aims to provide quick detection and automated responses to keep up with complex cyber threats in cloud environments.
Key facts
- Proposed a Deep Q-Network (DQN) for cloud intrusion detection
- Used CICIDS2017 dataset for training and UNSW-NB15 for validation
- Achieved 99.72% accuracy, 99.68% precision, 99.65% recall, 99.66% F1-score, ROC-AUC 0.999
- False positive rate 0.31%, false negative rate 0.35%, detection latency 15 ms
- Compared with decision tree, SVM, random forest, XGBoost, and multilayer perceptron
- Framework supports real-time detection and autonomous response
- Published on arXiv with ID 2608.12190
Entities
Institutions
- arXiv