Title:
ADAPTIVE CYBERSECURITY MONITORING VIA SEMANTIC STREAM PROCESSING AND GNN-BASED TRUST SCORING ON IPV4 LOGS

Authors:
Venkataramesh Induru and G. Arulkumaran, USA

Abstract:
Cybersecurity threats, including dynamic IP behaviour and device spoofing, operate as nitro boosts that pose considerable majorettes for real-time fraud detection systems. The existing models in practice seem too inflexible to adapt to the changing face of the fraud landscape, are not semanticaware, and fall short of graph-based reasoning. In this research, we propose an adaptive cybersecurity monitor whereby trust scores are combined over enriched IPv4 logs based on Graph Neural Networks (GNNs) with semantic stream processing. Complex Event Processing (CEP) and GNNs with exponentially time-decayed trust modelling are applied for semantic enrichment and verification of real-time data from transaction logs, SIEM alerts, and IP reputation feeds. Our method reduces false positives by 37 percent with a precision of 94.3% and an F1 score of 92.1%. Due to fast detection with low latency (MTTD: 1.4 seconds, MTTR: 2.7 seconds), the system is well suited for high-security applications such as office access control and online banking.