Automatic Learning coupled with Interpretability: MBDA in Action
Document Type
Conference Paper
Publication Date
6-2020
Publication Title
Proceedings of the Network Traffic Measurement and Analysis Conference (TMA)
Department
Department of Computer Science
Abstract
In this paper, we illustrate the application of Multivariate Big Data Analysis (MBDA), a recently proposed interpretable machine-learning method with application to Big Data sets. We apply MBDA for the first time for the detection and troubleshooting of network problems in a campus-wide Wi-Fi network. Data includes a seven-year trace (from 2012 to 2018) of the network’s most recent activity, with approximately 3,000 distinct access points, 40,000 authenticated users, and 600,000 distinct Wi-Fi stations. This is the longest and largest Wi-Fi trace known to date. Furthermore, we propose a new feature-learning procedure that solves an inherent limitation in MBDA: the manual definition of the features. The extended MBDA results in a methodology that allows network analysts to identify problems and diagnose them, which are principal tasks to troubleshoot the network and optimize its performance. In the paper, we go through the entire workflow of the approach, illustrating its application in detail and discussing processing times.
Original Citation
José Camacho, Rasmus Bro, and David Kotz. Automatic Learning coupled with Interpretability: MBDA in Action.Proceedings of the Network Traffic Measurement and Analysis Conference (TMA). IFIP, June 2020. ISBN13: 978-3-903176-27-0. ©Copyright European Union.
Dartmouth Digital Commons Citation
Camacho, José; Bro, Rasmus; and Kotz, David, "Automatic Learning coupled with Interpretability: MBDA in Action" (2020). Dartmouth Scholarship. 4030.
https://digitalcommons.dartmouth.edu/facoa/4030
