NeuralGuard: A Real-Time AI-Driven Autonomous Intrusion Prevention System for Network Security

Modern network security struggles with the “Response Gap” – the critical delay between threat detection and mitigation allowing sophisticated attacks like DDoS and botnets to succeed within seconds. This paper presents NeuralGuard, an autonomous Intrusion Prevention System achieving sub – 10ms threat response through hybrid deep learning and kernel-level firewall integration. The system employs dual-model architecture: a Multi-Layer Perceptron for supervised detection of known attacks and a deep autoencoder for unsupervised anomaly detection of zero-day threats. A Modified Genetic Algorithm optimizes features from 79 to 31 variables, reducing computational overhead by 66% while maintaining accuracy. Network flows are captured via Scapy, processed through 31 statistical features, and evaluated with 9.1ms end-to-end latency, enabling immediate automated firewall blocking via UFW/iptables/ipset. Evaluated on CIC-IDS-2017, NeuralGuard achieves 99.9% classification accuracy across 14 attack categories with an F1-score of 0.978 and false positive rate of 3 – 5%. A self-healing TTL mechanism (5-minute auto-recovery) prevents permanent blocking of legitimate users. Real-time Grafana dashboards and PostgreSQL logging provide forensic visibility. This work demonstrates that AI-orchestrated autonomous defense is practical, scalable, and deployable on resource-constrained systems, advancing autonomous network security.