Abstract
Abstract - This paper presents an AI-driven firewall that enhances network security by combining traditional traffic filtering with machine-learning–based threat detection. The system analyzes packet and flow features to identify anomalous behavior, classify malicious patterns, and enforce adaptive access-control decisions in real time. A Python-based inference pipeline performs data preprocessing, feature extraction, and model execution, while a JavaScript web interface provides policy management, alerts, and visualization for administrators. The proposed design supports continuous learning from observed traffic, reducing false positives and improving detection of zero-day and low-and-slow attacks. Experiments using representative benign and attack traffic demonstrate improved detection accuracy and response latency suitable for practical deployment, with clear audit logs for incident investigation. Keyword - AI-driven firewall, network security, intrusion detection, anomaly detection, machine learning, traffic classification, access control, real-time monitoring, zero-day attacks, web dashboard.