| Abstract : | The data obtained from the Gravitational Wave (GW) detectors is non-stationary and is plagued
with non-Gaussian noise, which often contains spurious noise transients (glitches). The change in
nature of these glitches as the sensitivity of the detectors changes makes the task of detecting transient GW signals in a noise plagued data even more challenging. For a given dataset, transient
gravitational wave searches produce a corresponding list of triggers corresponding to a gravitational
wave signal. These triggers could often result from glitches mimicking gravitational wave characteristics if the transient is short in nature, affecting the sensitivity of the search for systems such as
short-duration bursts or signals from very massive compact binaries. To distinguish glitches from
GW signals, search algorithms like coherent WaveBurst apply thresholds on the trigger properties.
Here, we present the Gaussian Mixture Models, a supervised machine learning approach, as a means
of modeling the multi-dimensional trigger attribute space. We fine-tune and optimize the model for
Intermediate Mass Black Hole (IMBH) detection and show an improvement in sensitivity, especially
towards high mass and high mass ratio binary systems. |
