Network Service Identificationvia Ipfix And Heterogeneousgraph

Sammanfattning

This research addresses the “service ambiguity” problem in encrypted network trafficclassification by moving beyond traditional Layer 3 and Layer 4 metrics. Utilizingthe Internet Microscope framework, this study implements a collection and learning pipeline aimed at service-level identification in shared IP infrastructures. Mov-ing away from deep packet inspection and application-layer encryption hints, theproposed methodology leverages hybrid weak supervision labelling function. Byembedding Layer 2 device context, service extraction via DNS logs, and enrichedflow telemetry into a heterogeneous graph topology, the methodology aims to over-come the predictive ceiling of flat tabular baselines and improve service-level dis-crimination under realistic household conditions The empirical dataset combinesone benchmark household with longer naturalistic and shorter guided householddeployments to balance high-confidence supervision with realistic domestic trafficVariability.