This project presents an innovative approach to enhancing network security through the integration of machine learning (ML) with eBPF-based packet filtering. Leveraging eXpress Data Path (XDP) technology, our solution enables the early detection and filtering of malicious packets at the network interface level, significantly reducing processing time and CPU overhead. We propose a novel method to incorporate ML models with floating-point weights into eBPF programs, overcoming the limitations imposed by eBPF's lack of support for floating-point arithmetic. Our design consists of an eBPF-based firewall attached to a network interface via XDP, which filters incoming packets based on user-defined rules and predictions from a Logistic Regression Model identifying blacklisted IP addresses. Evaluation results demonstrate the superior throughput performance of our kernel space firewall compared to traditional user-space firewalls. This research opens avenues for optimizing eBPF support for advanced ML techniques and underscores the importance of early detection in fortifying network security against evolving threats.