Performance study of Boosted Decision Trees in a Gravitational Wave Burst Search.

Document #:

LIGO-G0900565-v7

Document type:

G - Presentations (eg Graphics)

Abstract:

We explore the implementation of multivariate classification analysis in gravitational wave burst searches, to separate signals from background.
We focus on the Boosted Decision Tree algorithm and the coincidence between three interferometers (two of which co-located) as applied in the LSC S5 burst analysis with the Omega pipeline. The Boosted Decision Tree algorithm, from the ROOT TMVA package, is applied to bandwidth, duration, H1H2 coherent energy, H1H2 correlated energy and L1 normalized energy. In this preliminary study, the signal sample are simulated binary black hole coalescences (EOBNR waveform, coded in LALApps, in the total mass range 100-350 solar mass) and the background sample are accidental coincidences in simulated gaussian noise. We compare the signal-noise discrimination obtained by the BDT algorithm to the cuts applied in the LSC burst analysis of the first year of S5.

Files in Document:

• poster.pdf (724.2 kB)

Topics:

• GW Bursts

Author Groups:

• LSC