Mining Conceptual Knowledge from Network Traffic Data for Traffic Measurement Optimization

Formal concept analysis (FCA) is a knowledge discovery approach aimed at extracting conceptual hierarchies from data. Due to the exhaustiveness of its output, a typical FCA-based solution would filter large parts thereof using an ad-hoc quality criterion. In this paper, we present an FCA-based solution to an optimization problem from network traffic control that is akin to information retrieval (queries set up to measure specific traffic, i.e., packet flows). The goal is to minimize the number of counters used to answer a given query set. Our solution explores a contextual substructure of the (flows x flow descriptors) lattice, that we called the projection subsemilattice: The optimal set of counters is shown to correspond to a class of concepts from the semilattice. We present an effective computing method and provide empirical evidence of its performances on realistic network settings.