Evaluating Deep Learning Models for Multiclass Classification of LIGO Gravitational Wave Glitches

Gravitational-wave observatories such as LIGO have revolutionized astrophysics by detecting ripples in spacetime from cataclysmic events. However, transient noise artifacts, known as glitches, remain a significant challenge by mimicking or obscuring signals. Accurate classification of these glitches is essential for improving the fidelity of gravitational-wave analysis. Using LIGO data, we investigate 24 glitch classes in a multiclass classification framework. We present a comparative evaluation of traditional Gradient Boosted Trees (GBT) and deep learning models applied to tabular glitch features. Our results confirm that while GBT remains a strong baseline, deep learning models achieve competitive F1 scores when optimized with preprocessing, feature engineering, and imbalance-aware strategies. This work underscores the importance of tailored ML approaches for noise characterization in gravitational-wave detectors and highlights pathways to improve data quality, thereby enabling more reliable astrophysical inference from future gravitational-wave detections.