Machine learning techniques for gravitational waves data analysis - random forest - improving MBTA signal-noise discrimination

Aug 2, 2026

Machine learning techniques are increasingly used to study gravitational-wave physics and to strengthen the separation of true signals from detector noise. The work presents a random-forest-based tool that improves the MBTA (Multi-Band Template Analysis) algorithm for online compact-binary-coalescence analysis. Multiple configurations and feature sets—combining physical and statistical trigger quantities—are used to train and test the classifier. Real data are then used to compare the statistical significance from the machine-learning approach with the classical MBTA pipeline results.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

Introduction

MBTA

Multi-Band Template Analysis pipeline

Machine learning for gravitational waves

Random forest algorithm

IL NUOVO CIMENTO 48 C (2025) 99 DOI 10.1393/ncc/i2025-25099-8 Colloquia: IFAE 2024 Machine learning techniques for gravitational waves data analysis (∗ ) L. Mobilia ( 1 )(2 )(∗∗ ) and the MBTA Collaboration (1 ) Sez. di Fisica, Universit`a degli Studi di Urbino Carlo Bo - Urbino, Italy (2 ) INFN, Sezione di Firenze - Firenze, Italy received 2 December 2024 Summary.— The use of machine learning in the study of gravitational wave physics is increasingly widespread. The ﬂexibility and results that this technology has achieved encourage the use and exploration of such techniques in this research ﬁeld. In this work, we develop a machine learning tool based on the random forest technique to enhance the measurement capabilities of the MBTA (Multi-Band Template Analysis) algorithm in distinguishing signal from noise. The results are obtained by considering diﬀerent conﬁgurations and features, taking into account both physical and statistical values of the triggers to train and test the machine learning algorithm. Comparisons between the statistical signiﬁcance obtained from machine learning and the classical algorithm were conducted using real data. 1.– Introduction The ﬁrst detection of a gravitational wave signal [3] obtained by the LIGO interferometers opened the door to the gravitational astronomy [1] . Since then, more than 90 events have been measured [6], [7], [8], [9] by the LIGO-Virgo-Kagra collaboration. After the discovery of the ﬁrst binary neutron star system [4], the multi-messenger gravitational wave astronomy has become a reality, with important consequences for cosmological studies and new physics [2], [5] . Those results have been achieved through the interferometers LIGO [10] in the United States of America and Virgo [11] in Italy. In order to detect the gravitational waves signals several algorithms have been developed, for both modelled and un-modelled search [15] . For the compact binary coalescence (CBC) online analysis, that consists in the search of astrophysical compact objects such as binary black holes, a particular type of pipelines that relies on the matched-ﬁltering method are required [12], [13], [14] . In this contribution, we will try to increase the detection capability of the pipeline MBTA, a CBC pipeline currently used for the online analysis, by adopting machine learning techniques. (∗ ) IFAE 2024- “Cosmology and Astroparticles” session ( ∗∗ ) E-mail: [l.mobilia@campus.uniurb.it](l.mobilia@campus.uniurb.it) Creative Commons Attribution 4.0 License ([https://creativecommons.org/licenses/by/4.0](https://creativecommons.org/licenses/by/4.0)) 1 2 L. MOBILIA and the MBTA COLLABORATION Fig. 1.– Scheme of the MBTA pipeline: matched-ﬁltering for each band is evaluated from the detectors’ stream then, if a trigger occurs, the false alarm rate is evaluated and the event is uploaded to the database. 2.– MBTA The Multi-Band Template Analysis (MBTA) pipeline is an algorithm based on the matched-ﬁltering technique for the compact binary coalescence events detection [16],[17] . In order to work, the matched-ﬁltering methods requires a ’bank’ of templates that is built with an hybrid-code [18] . MBTA analyses the data stream provided by the output of the interferometers considering chunks of several seconds each. In each chunk the matched-ﬁltering is applied. In MBTA, this procedure is done independently and in parallel for both low and high frequencies, dividing accordingly the template bank and so reducing considerably the computational time. If the matched-ﬁltering analysis results ina signal to noise ratio value above a certain threshold for both the frequency bands, then a triggers is produced. Each trigger has several parameters’ such as the signal to noise ratio ρ, the χ2 or the masses and the spins. The matched-ﬁltering technique is optimal incase of Gaussian noise, but for real noise data, the presence of glitches must be taken into account. In the actual conﬁguration, MBTA

FAQs

What problem does the document address in gravitational-wave detection?

It addresses how to improve the MBTA pipeline’s capability to distinguish real gravitational-wave signals from noise and detector glitches during compact binary coalescence searches.

How is the random forest used to enhance MBTA?

A random-forest model is trained using different trigger feature configurations, including both physical and statistical values, to improve separation between noise and signal.

How are the machine-learning results validated against the classical approach?

The work compares statistical significance obtained from the random-forest-based method with the classical MBTA algorithm using real data, and uses injection-based testing to ensure robust statistics.

Aug 2, 2026

Upload by River Wang