Activity Number:

315

Type:

Contributed

Date/Time:

Tuesday, August 2, 2016 : 8:30 AM to 10:20 AM

Sponsor:

Section on Statistical Learning and Data Science

Abstract #318698

View Presentation

Title:

Hypothesis Testing for Detecting Changes Within a BarabásiAlbert Network

Author(s):

Fairul MohdZaid* and Christine Schubert Kabban and Edward White and Richard Deckro

Companies:

Air Force Research Lab and Air Force Institute of Technology and Air Force Institute of Technology and Air Force Institute of Technology

Keywords:

BarabásiAlbert ;
Classification ;
Hypothesis Test ;
Power Law ;
Networks ;
Lmoments

Abstract:

A hypothesis test for detecting changes within the BarabásiAlbert network is proposed. We characterize change as a proportion of edge deletion based upon the degree distribution for the nodes within the network. The degree distribution of the network is assumed to follow the Pareto distribution. Therefore, a statistical test based on the joint distribution of multiple Lmoments for the Pareto distribution is developed. Simulation results examined the power of this test for various network sizes and proportion of edge deletion. Sensitivity of the proportion of deleted edges to nodal degree was investigated using general categories of small, medium, and large degree nodes. These categories were defined by percentiles of the degree distribution for each network size examined. Results demonstrate that this test has high power for detecting changes even for small networks (32 nodes) and when a small proportion of edges are deleted (e.g. 0.01). Implications for these results suggest that this test may be utilized for early detection of changes occurring within the BarabásiAlbert network model which may be too subtle for tests based upon lower moments such as the mean.
