Advanced WIreless Network Activity Inference: Monitoring wireless communivation meta data.
Translated title
Advanced WIreless Network Activity Inference
Author
Toft, Christian Hilligsøe
Term
4. semester
Education
Publication year
2020
Abstract
Dette speciale undersøger, om applikationer kan identificeres ud fra offentligt tilgængelige Wi-Fi-metadata – altså forbindelsesoplysninger og ikke selve indholdet af trafikken. Vi indsamlede Wi-Fi-trafik på Aalborg Universitet ved passiv overvågning og forsøgte at gruppere trafikken efter applikation med en usuperviseret klyngemetode (en Gaussian mixture model). Før klyngedannelsen anvendte vi tre forbehandlingsstrategier: simpel standardisering; Factor Analysis of Mixed Data (en metode til at reducere mange variable til et mindre sæt); og en neuralt netværks-baseret kodning via en Gated Recurrent Unit (GRU) autoencoder, som komprimerer sekvenser. Med disse features og evaluering på data med begrænset mærkning klarede klyngerne sig ikke bedre end tilfældig gætning til at identificere applikationer i dette datasæt.
This thesis investigates whether applications can be identified from publicly available Wi-Fi metadata—connection information rather than the content of the traffic. We passively recorded Wi-Fi traffic at Aalborg University and used an unsupervised clustering method (a Gaussian mixture model) to group traffic with the goal of separating applications. Before clustering, we applied three preprocessing strategies: simple standardization; Factor Analysis of Mixed Data (a technique that reduces many variables to a smaller set); and a neural network encoder based on a Gated Recurrent Unit (GRU) autoencoder that compresses sequences. Using these features and evaluating on data with restricted labeling, the clustering did not perform better than random guessing at identifying applications in this dataset.
[This abstract was generated with the help of AI]
Keywords
Documents